tedit/app_image.h
2025-04-07 20:08:16 -04:00

1759 lines
70 KiB
C++

#ifndef APP_IMAGE_H
#define APP_IMAGE_H
#include <vector>
#include <string>
#include <map>
#include <optional> // Requires C++17
#include <memory>
#include <cstdint>
#include <cmath>
#include <fstream>
#include <stdexcept>
#include <algorithm>
#include <iostream> // For errors/warnings
#include <cstring> // For strcmp, memcpy, etc.
#include <setjmp.h> // For libjpeg/libpng error handling
#define IMGUI_DEFINE_MATH_OPERATORS // Allows ImVec2 operators
#include "imgui_internal.h" // Need ImFloorSigned, ImClamp, ImMax, ImMin, ImAbs
// --- User Instructions ---
// 1. Place easyexif.h in your include path.
// 2. Ensure development libraries for LibRaw, libjpeg-turbo, libpng, and libtiff are installed.
// 3. In EXACTLY ONE .cpp file in your project, before including this header, define:
// #define APP_IMAGE_IMPLEMENTATION
// 4. When compiling, LINK against the necessary libraries, e.g., using CMake or directly:
// g++ your_app.cpp -o your_app -std=c++17 -lraw -ljpeg -lpng -ltiff -lm (order might matter)
// --- Forward declarations of external library types (optional, mostly for clarity) ---
// struct jpeg_decompress_struct;
// struct jpeg_compress_struct;
// struct jpeg_error_mgr;
// struct png_struct_def;
// struct png_info_def;
// typedef struct tiff TIFF; // From tiffio.h
// class LibRaw; // From libraw/libraw.h
// Include easyexif here as it's header-only anyway
#include "exif.h"
// Enum for specifying save formats
enum class ImageSaveFormat
{
JPEG, // Quality setting applies (1-100), saves as 8-bit sRGB.
PNG_8, // 8-bit PNG (sRGB assumption).
PNG_16, // 16-bit PNG (Linear or sRGB depends on future implementation details, currently linear).
TIFF_8, // 8-bit TIFF (Uncompressed, RGB).
TIFF_16, // 16-bit TIFF (Uncompressed, RGB, Linear).
// TIFF_LZW_16 // Example for compressed TIFF
UNKNOWN
};
// Basic structure for image metadata (can hold EXIF tags)
using ImageMetadata = std::map<std::string, std::string>;
// --- App Internal Image Representation ---
class AppImage
{
public:
// --- Constructors ---
AppImage() = default;
AppImage(uint32_t width, uint32_t height, uint32_t channels = 3);
// --- Accessors ---
uint32_t getWidth() const { return m_width; }
uint32_t getHeight() const { return m_height; }
uint32_t getChannels() const { return m_channels; }
bool isEmpty() const { return m_pixelData.empty(); }
// Pixel data: Linear floating point [0.0, 1.0+], interleaved RGB/RGBA/Gray.
float *getData() { return m_pixelData.data(); }
const float *getData() const { return m_pixelData.data(); }
size_t getDataSize() const { return m_pixelData.size() * sizeof(float); }
size_t getTotalFloats() const { return m_pixelData.size(); }
std::vector<float> &getPixelVector() { return m_pixelData; }
const std::vector<float> &getPixelVector() const { return m_pixelData; }
// --- Metadata ---
ImageMetadata &getMetadata() { return m_metadata; }
const ImageMetadata &getMetadata() const { return m_metadata; }
// --- Color Information ---
std::vector<uint8_t> &getIccProfile() { return m_iccProfile; }
const std::vector<uint8_t> &getIccProfile() const { return m_iccProfile; }
std::string &getColorSpaceName() { return m_colorSpaceName; }
const std::string &getColorSpaceName() const { return m_colorSpaceName; }
bool isLinear() const { return m_isLinear; }
// --- Modifiers ---
void resize(uint32_t newWidth, uint32_t newHeight, uint32_t newChannels = 0);
void clear_image();
// --- Data members ---
// Making them public for easier access in the implementation section below,
// alternatively make loadImage/saveImage friends or add internal setters.
// public:
uint32_t m_width = 0;
uint32_t m_height = 0;
uint32_t m_channels = 0; // 1=Gray, 3=RGB, 4=RGBA
std::vector<float> m_pixelData;
ImageMetadata m_metadata;
std::vector<uint8_t> m_iccProfile;
std::string m_colorSpaceName = "Unknown";
bool m_isLinear = true; // Default assumption for internal format
GLuint m_textureId = 0;
int m_textureWidth = 0;
int m_textureHeight = 0;
};
// --- API Function Declarations ---
/**
* @brief Loads an image file, attempting type detection (RAW, JPEG, PNG, TIFF).
* Uses LibRaw, libjpeg-turbo, libpng, libtiff.
* Uses EasyExif for EXIF metadata from JPEGs (only).
* Converts loaded pixel data to internal linear float format.
* Extracts ICC profile if available (primarily from RAW).
*
* @param filePath Path to the image file.
* @return std::optional<AppImage> containing the loaded image on success, std::nullopt on failure.
*/
std::optional<AppImage> loadImage(const std::string &filePath);
/**
* @brief Saves the AppImage to a file (JPEG, PNG, TIFF).
* Uses libjpeg-turbo, libpng, libtiff.
* Converts internal linear float data to target format (e.g., 8-bit sRGB for JPEG).
* NOTE: Does NOT currently save EXIF or ICC metadata. This requires more complex handling
* (e.g., using Exiv2 library or manual file manipulation after saving pixels).
*
* @param image The AppImage to save. Assumed to be in linear float format.
* @param filePath Path to save the image file.
* @param format The desired output format.
* @param quality JPEG quality (1-100), ignored otherwise.
* @return True on success, false on failure.
*/
bool saveImage(const AppImage &image,
const std::string &filePath,
ImageSaveFormat format,
int quality = 90);
bool loadImageTexture(const AppImage &appImage);
// ============================================================================
// =================== IMPLEMENTATION SECTION =================================
// ============================================================================
// Define APP_IMAGE_IMPLEMENTATION in exactly one .cpp file before including this header
#ifdef APP_IMAGE_IMPLEMENTATION
#include <libraw/libraw.h>
#include <jpeglib.h>
#include <png.h>
#include <tiffio.h>
// Internal helper namespace
namespace AppImageUtil
{
// --- Error Handling ---
// Basic error reporting (prints to stderr)
inline void LogError(const std::string &msg)
{
std::cerr << "AppImage Error: " << msg << std::endl;
}
inline void LogWarning(const std::string &msg)
{
std::cerr << "AppImage Warning: " << msg << std::endl;
}
// --- Color Conversion Helpers (Approximate sRGB) ---
// For critical work, use a color management library (LittleCMS) and proper piecewise functions
inline float srgb_to_linear_approx(float srgbVal)
{
if (srgbVal <= 0.0f)
return 0.0f;
if (srgbVal <= 0.04045f)
{
return srgbVal / 12.92f;
}
else
{
return std::pow((srgbVal + 0.055f) / 1.055f, 2.4f);
}
}
inline float linear_to_srgb_approx(float linearVal)
{
if (linearVal <= 0.0f)
return 0.0f;
// Simple clamp for typical display output
linearVal = std::fmax(0.0f, std::fmin(1.0f, linearVal));
if (linearVal <= 0.0031308f)
{
return linearVal * 12.92f;
}
else
{
return 1.055f * std::pow(linearVal, 1.0f / 2.4f) - 0.055f;
}
}
// --- File Type Detection ---
enum class DetectedFileType
{
RAW,
JPEG,
PNG,
TIFF,
UNKNOWN
};
inline DetectedFileType detectFileType(const std::string &filePath)
{
std::ifstream file(filePath, std::ios::binary);
if (!file)
return DetectedFileType::UNKNOWN;
unsigned char magic[12]; // Read enough bytes for common signatures
file.read(reinterpret_cast<char *>(magic), sizeof(magic));
if (!file)
return DetectedFileType::UNKNOWN;
// Check common signatures
if (magic[0] == 0xFF && magic[1] == 0xD8 && magic[2] == 0xFF)
return DetectedFileType::JPEG;
if (magic[0] == 0x89 && magic[1] == 'P' && magic[2] == 'N' && magic[3] == 'G')
return DetectedFileType::PNG;
if ((magic[0] == 'I' && magic[1] == 'I' && magic[2] == 0x2A && magic[3] == 0x00) || // Little-endian TIFF
(magic[0] == 'M' && magic[1] == 'M' && magic[2] == 0x00 && magic[3] == 0x2A)) // Big-endian TIFF
{
size_t dotPos = filePath.rfind('.');
if (dotPos != std::string::npos)
{
std::string ext = filePath.substr(dotPos);
std::transform(ext.begin(), ext.end(), ext.begin(), ::tolower);
// Common RAW formats that use TIFF structure
const char *rawTiffExtensions[] = {
".nef", // Nikon
".cr2", // Canon
".dng", // Adobe/Various
".arw", // Sony
".srw", // Samsung
".orf", // Olympus
".pef", // Pentax
".raf", // Fuji
".rw2" // Panasonic
};
for (const char *rawExt : rawTiffExtensions)
{
if (ext == rawExt)
return DetectedFileType::RAW;
}
}
return DetectedFileType::TIFF;
}
// If no standard signature matches, check extension for RAW as a fallback
// (LibRaw handles many internal variations)
size_t dotPos = filePath.rfind('.');
if (dotPos != std::string::npos)
{
std::string ext = filePath.substr(dotPos);
std::transform(ext.begin(), ext.end(), ext.begin(), ::tolower);
const char *rawExtensions[] = {
".3fr", ".ari", ".arw", ".bay", ".braw", ".crw", ".cr2", ".cr3", ".cap",
".data", ".dcs", ".dcr", ".dng", ".drf", ".eip", ".erf", ".fff", ".gpr",
".iiq", ".k25", ".kdc", ".mdc", ".mef", ".mos", ".mrw", ".nef", ".nrw",
".obm", ".orf", ".pef", ".ptx", ".pxn", ".r3d", ".raf", ".raw", ".rwl",
".rw2", ".rwz", ".sr2", ".srf", ".srw", ".tif", ".x3f" // Note: .tif can be RAW or regular TIFF
};
for (const char *rawExt : rawExtensions)
{
if (ext == rawExt)
return DetectedFileType::RAW;
}
// Special case: Leica .dng can also be loaded by LibRaw
if (ext == ".dng")
return DetectedFileType::RAW;
}
return DetectedFileType::UNKNOWN;
}
// --- EXIF Loading Helper (using EasyExif) ---
inline void loadExifData(const std::string &filePath, ImageMetadata &metadata)
{
std::ifstream file(filePath, std::ios::binary | std::ios::ate);
if (!file)
return;
std::streamsize size = file.tellg();
file.seekg(0, std::ios::beg);
std::vector<unsigned char> buffer(size);
if (!file.read(reinterpret_cast<char *>(buffer.data()), size))
return;
easyexif::EXIFInfo exifInfo;
int code = exifInfo.parseFrom(buffer.data(), buffer.size());
if (code == 0)
{
// Helper lambda to add if not empty
auto addMeta = [&](const std::string &key, const std::string &value)
{
if (!value.empty())
metadata[key] = value;
};
auto addMetaInt = [&](const std::string &key, int value)
{
if (value > 0)
metadata[key] = std::to_string(value);
};
auto addMetaDouble = [&](const std::string &key, double value)
{
if (value > 0)
metadata[key] = std::to_string(value);
};
addMeta("Exif.Image.Make", exifInfo.Make);
addMeta("Exif.Image.Model", exifInfo.Model);
addMeta("Exif.Image.Software", exifInfo.Software);
addMetaInt("Exif.Image.Orientation", exifInfo.Orientation);
addMeta("Exif.Image.DateTime", exifInfo.DateTime);
addMeta("Exif.Photo.DateTimeOriginal", exifInfo.DateTimeOriginal);
addMeta("Exif.Photo.DateTimeDigitized", exifInfo.DateTimeDigitized);
addMeta("Exif.Image.SubSecTimeOriginal", exifInfo.SubSecTimeOriginal); // Often empty
addMeta("Exif.Image.Copyright", exifInfo.Copyright);
addMetaDouble("Exif.Photo.ExposureTime", exifInfo.ExposureTime);
addMetaDouble("Exif.Photo.FNumber", exifInfo.FNumber);
addMetaInt("Exif.Photo.ISOSpeedRatings", exifInfo.ISOSpeedRatings);
addMetaDouble("Exif.Photo.ShutterSpeedValue", exifInfo.ShutterSpeedValue); // APEX
addMetaDouble("Exif.Photo.ApertureValue", exifInfo.FNumber); // APEX
addMetaDouble("Exif.Photo.ExposureBiasValue", exifInfo.ExposureBiasValue);
addMetaDouble("Exif.Photo.FocalLength", exifInfo.FocalLength);
addMeta("Exif.Photo.LensModel", exifInfo.LensInfo.Model);
// GeoLocation
if (exifInfo.GeoLocation.Latitude != 0 || exifInfo.GeoLocation.Longitude != 0)
{
metadata["Exif.GPSInfo.Latitude"] = std::to_string(exifInfo.GeoLocation.Latitude);
metadata["Exif.GPSInfo.Longitude"] = std::to_string(exifInfo.GeoLocation.Longitude);
metadata["Exif.GPSInfo.Altitude"] = std::to_string(exifInfo.GeoLocation.Altitude);
metadata["Exif.GPSInfo.LatitudeRef"] = exifInfo.GeoLocation.LatComponents.direction;
metadata["Exif.GPSInfo.LongitudeRef"] = exifInfo.GeoLocation.LonComponents.direction;
}
}
else
{
// LogWarning("Could not parse EXIF data (Code " + std::to_string(code) + ") from " + filePath);
}
}
// --- LibRaw Loading ---
inline std::optional<AppImage> loadRaw(const std::string &filePath)
{
LibRaw rawProcessor;
AppImage image;
// Set parameters for desired output
// Output 16-bit data
rawProcessor.imgdata.params.output_bps = 16;
// Disable automatic brightness adjustment (we want linear)
rawProcessor.imgdata.params.no_auto_bright = 1;
// Set output color space (e.g., 1 = sRGB, 3 = ProPhoto, 4 = AdobeRGB)
// ProPhoto (3) or AdobeRGB (4) are good wide-gamut choices if editor supports them.
// sRGB (1) is safest if unsure. We'll assume Linear sRGB for now.
rawProcessor.imgdata.params.output_color = 1; // 1 = sRGB primaries
// Set gamma (1.0 for linear) - use {1.0, 1.0} for linear output
rawProcessor.imgdata.params.gamm[0] = 1.0; // Linear gamma
rawProcessor.imgdata.params.gamm[1] = 1.0;
// Use camera white balance if available, otherwise auto
rawProcessor.imgdata.params.use_camera_wb = 1;
rawProcessor.imgdata.params.use_auto_wb = (rawProcessor.imgdata.params.use_camera_wb == 0);
// Consider other params: demosaic algorithm, highlight recovery, etc.
int ret;
if ((ret = rawProcessor.open_file(filePath.c_str())) != LIBRAW_SUCCESS)
{
LogError("LibRaw: Cannot open file " + filePath + " - " + libraw_strerror(ret));
return std::nullopt;
}
if ((ret = rawProcessor.unpack()) != LIBRAW_SUCCESS)
{
LogError("LibRaw: Cannot unpack file " + filePath + " - " + libraw_strerror(ret));
return std::nullopt;
}
// Process the image (demosaic, color conversion, etc.)
if ((ret = rawProcessor.dcraw_process()) != LIBRAW_SUCCESS)
{
LogError("LibRaw: Cannot process file " + filePath + " - " + libraw_strerror(ret));
// Try fallback processing if dcraw_process fails (might be non-RAW TIFF/JPEG)
if (ret == LIBRAW_UNSUPPORTED_THUMBNAIL || ret == LIBRAW_REQUEST_FOR_NONEXISTENT_IMAGE)
{
LogWarning("LibRaw: File " + filePath + " might be non-RAW or only has thumbnail. Attempting fallback.");
// You could try loading with libjpeg/libtiff here, but for simplicity we fail
}
return std::nullopt;
}
// Get the processed image data
libraw_processed_image_t *processed_image = rawProcessor.dcraw_make_mem_image(&ret);
if (!processed_image)
{
LogError("LibRaw: Cannot make memory image for " + filePath + " - " + libraw_strerror(ret));
return std::nullopt;
}
// Copy data to AppImage format
if (processed_image->type == LIBRAW_IMAGE_BITMAP && processed_image->bits == 16)
{
image.m_width = processed_image->width;
image.m_height = processed_image->height;
image.m_channels = processed_image->colors; // Should be 3 (RGB)
image.m_isLinear = true; // We requested linear gamma
if (image.m_channels != 3)
{
LogWarning("LibRaw: Expected 3 channels, got " + std::to_string(image.m_channels));
// Handle grayscale or other cases if needed, for now assume RGB
image.m_channels = 3;
}
size_t num_pixels = static_cast<size_t>(image.m_width) * image.m_height;
size_t total_floats = num_pixels * image.m_channels;
image.m_pixelData.resize(total_floats);
uint16_t *raw_data = reinterpret_cast<uint16_t *>(processed_image->data);
float *app_data = image.m_pixelData.data();
// Convert 16-bit unsigned short [0, 65535] to float [0.0, 1.0+]
for (size_t i = 0; i < total_floats; ++i)
{
app_data[i] = static_cast<float>(raw_data[i]) / 65535.0f;
}
// Get color space name based on output_color param
switch (rawProcessor.imgdata.params.output_color)
{
case 1:
image.m_colorSpaceName = "Linear sRGB";
break;
case 2:
image.m_colorSpaceName = "Linear Adobe RGB (1998)";
break; // Check LibRaw docs if this is correct mapping
case 3:
image.m_colorSpaceName = "Linear ProPhoto RGB";
break;
case 4:
image.m_colorSpaceName = "Linear XYZ";
break; // Check LibRaw docs
default:
image.m_colorSpaceName = "Linear Unknown";
break;
}
// Extract Metadata (Example - add more fields as needed)
image.m_metadata["LibRaw.Camera.Make"] = rawProcessor.imgdata.idata.make;
image.m_metadata["LibRaw.Camera.Model"] = rawProcessor.imgdata.idata.model;
image.m_metadata["LibRaw.Image.Timestamp"] = std::to_string(rawProcessor.imgdata.other.timestamp);
image.m_metadata["LibRaw.Image.ShotOrder"] = std::to_string(rawProcessor.imgdata.other.shot_order);
image.m_metadata["LibRaw.Photo.ExposureTime"] = std::to_string(rawProcessor.imgdata.other.shutter);
image.m_metadata["LibRaw.Photo.Aperture"] = std::to_string(rawProcessor.imgdata.other.aperture);
image.m_metadata["LibRaw.Photo.ISOSpeed"] = std::to_string(rawProcessor.imgdata.other.iso_speed);
image.m_metadata["LibRaw.Photo.FocalLength"] = std::to_string(rawProcessor.imgdata.other.focal_len);
// Copy EasyExif compatible fields if possible for consistency
image.m_metadata["Exif.Image.Make"] = rawProcessor.imgdata.idata.make;
image.m_metadata["Exif.Image.Model"] = rawProcessor.imgdata.idata.model;
image.m_metadata["Exif.Photo.ExposureTime"] = std::to_string(rawProcessor.imgdata.other.shutter);
image.m_metadata["Exif.Photo.FNumber"] = std::to_string(rawProcessor.imgdata.other.aperture); // Aperture == FNumber
image.m_metadata["Exif.Photo.ISOSpeedRatings"] = std::to_string(rawProcessor.imgdata.other.iso_speed);
image.m_metadata["Exif.Photo.FocalLength"] = std::to_string(rawProcessor.imgdata.other.focal_len);
// LibRaw often provides DateTimeOriginal via timestamp
// Convert timestamp to string if needed:
// time_t ts = rawProcessor.imgdata.other.timestamp;
// char buf[30];
// strftime(buf, sizeof(buf), "%Y:%m:%d %H:%M:%S", localtime(&ts));
// image.m_metadata["Exif.Photo.DateTimeOriginal"] = buf;
// Extract ICC Profile
unsigned int icc_size = 0;
const void *icc_profile_ptr = nullptr;
if (icc_profile_ptr && icc_size > 0)
{
image.m_iccProfile.resize(icc_size);
std::memcpy(image.m_iccProfile.data(), icc_profile_ptr, icc_size);
LogWarning("LibRaw: Successfully extracted ICC profile (" + std::to_string(icc_size) + " bytes).");
// We could potentially parse the ICC profile name here, but it's complex.
if (image.m_colorSpaceName == "Linear Unknown")
image.m_colorSpaceName = "Linear (Embedded ICC)";
}
else
{
LogWarning("LibRaw: No ICC profile found or extracted.");
}
}
else
{
LogError("LibRaw: Processed image is not 16-bit bitmap (type=" + std::to_string(processed_image->type) + " bits=" + std::to_string(processed_image->bits) + ")");
LibRaw::dcraw_clear_mem(processed_image);
return std::nullopt;
}
// Clean up LibRaw resources
LibRaw::dcraw_clear_mem(processed_image);
// rawProcessor is automatically cleaned up by its destructor
return image;
}
// --- libjpeg Loading ---
// Custom error handler for libjpeg
struct JpegErrorManager
{
jpeg_error_mgr pub;
jmp_buf setjmp_buffer; // For returning control on error
};
void jpegErrorExit(j_common_ptr cinfo)
{
JpegErrorManager *myerr = reinterpret_cast<JpegErrorManager *>(cinfo->err);
// Format the error message
char buffer[JMSG_LENGTH_MAX];
(*cinfo->err->format_message)(cinfo, buffer);
LogError("libjpeg: " + std::string(buffer));
// Return control to setjmp point
longjmp(myerr->setjmp_buffer, 1);
}
inline std::optional<AppImage> loadJpeg(const std::string &filePath)
{
FILE *infile = fopen(filePath.c_str(), "rb");
if (!infile)
{
LogError("Cannot open JPEG file: " + filePath);
return std::nullopt;
}
AppImage image;
jpeg_decompress_struct cinfo;
JpegErrorManager jerr; // Custom error handler
// Setup error handling
cinfo.err = jpeg_std_error(&jerr.pub);
jerr.pub.error_exit = jpegErrorExit;
if (setjmp(jerr.setjmp_buffer))
{
// If we get here, a fatal error occurred
jpeg_destroy_decompress(&cinfo);
fclose(infile);
return std::nullopt;
}
// Initialize decompression object
jpeg_create_decompress(&cinfo);
jpeg_stdio_src(&cinfo, infile);
// Read header
jpeg_read_header(&cinfo, TRUE);
// Start decompressor - this guesses output parameters like color space
// We usually get JCS_RGB for color JPEGs
cinfo.out_color_space = JCS_RGB; // Request RGB output
jpeg_start_decompress(&cinfo);
image.m_width = cinfo.output_width;
image.m_height = cinfo.output_height;
image.m_channels = cinfo.output_components; // Should be 3 for JCS_RGB
if (image.m_channels != 1 && image.m_channels != 3)
{
LogError("libjpeg: Unsupported number of channels: " + std::to_string(image.m_channels));
jpeg_finish_decompress(&cinfo);
jpeg_destroy_decompress(&cinfo);
fclose(infile);
return std::nullopt;
}
size_t num_pixels = static_cast<size_t>(image.m_width) * image.m_height;
size_t total_floats = num_pixels * image.m_channels;
image.m_pixelData.resize(total_floats);
image.m_isLinear = true; // We will convert to linear
image.m_colorSpaceName = "Linear sRGB"; // Standard JPEG assumption
// Allocate temporary buffer for one scanline
int row_stride = cinfo.output_width * cinfo.output_components;
std::vector<unsigned char> scanline_buffer(row_stride);
JSAMPROW row_pointer[1];
row_pointer[0] = scanline_buffer.data();
float *app_data_ptr = image.m_pixelData.data();
// Read scanlines
while (cinfo.output_scanline < cinfo.output_height)
{
jpeg_read_scanlines(&cinfo, row_pointer, 1);
// Convert scanline from 8-bit sRGB to linear float
for (int i = 0; i < row_stride; ++i)
{
*app_data_ptr++ = srgb_to_linear_approx(static_cast<float>(scanline_buffer[i]) / 255.0f);
}
}
// Finish decompression and clean up
jpeg_finish_decompress(&cinfo);
jpeg_destroy_decompress(&cinfo);
fclose(infile);
// Load EXIF data separately
loadExifData(filePath, image.m_metadata);
return image;
}
// --- libpng Loading ---
// Custom error handler for libpng
void pngErrorFunc(png_structp png_ptr, png_const_charp error_msg)
{
LogError("libpng: " + std::string(error_msg));
jmp_buf *jmp_ptr = reinterpret_cast<jmp_buf *>(png_get_error_ptr(png_ptr));
if (jmp_ptr)
{
longjmp(*jmp_ptr, 1);
}
// If no jmp_buf, just exit (shouldn't happen if setup correctly)
exit(EXIT_FAILURE);
}
void pngWarningFunc(png_structp png_ptr, png_const_charp warning_msg)
{
LogWarning("libpng: " + std::string(warning_msg));
// Don't longjmp on warnings
}
inline std::optional<AppImage> loadPng(const std::string &filePath)
{
FILE *fp = fopen(filePath.c_str(), "rb");
if (!fp)
{
LogError("Cannot open PNG file: " + filePath);
return std::nullopt;
}
// Check PNG signature
unsigned char header[8];
fread(header, 1, 8, fp);
if (png_sig_cmp(header, 0, 8))
{
LogError("File is not a valid PNG: " + filePath);
fclose(fp);
return std::nullopt;
}
png_structp png_ptr = png_create_read_struct(PNG_LIBPNG_VER_STRING, nullptr, pngErrorFunc, pngWarningFunc);
if (!png_ptr)
{
LogError("libpng: png_create_read_struct failed");
fclose(fp);
return std::nullopt;
}
png_infop info_ptr = png_create_info_struct(png_ptr);
if (!info_ptr)
{
LogError("libpng: png_create_info_struct failed");
png_destroy_read_struct(&png_ptr, nullptr, nullptr);
fclose(fp);
return std::nullopt;
}
// Setup jump buffer for error handling
jmp_buf jmpbuf;
if (setjmp(jmpbuf))
{
LogError("libpng: Error during read");
png_destroy_read_struct(&png_ptr, &info_ptr, nullptr);
fclose(fp);
return std::nullopt;
}
// Assign jump buffer to png error pointer
// Note: The cast from jmp_buf* to png_voidp* might feel odd, but it's standard practice
png_set_error_fn(png_ptr, reinterpret_cast<png_voidp>(&jmpbuf), pngErrorFunc, pngWarningFunc);
png_init_io(png_ptr, fp);
png_set_sig_bytes(png_ptr, 8); // We already read the 8 signature bytes
// Read file info
png_read_info(png_ptr, info_ptr);
AppImage image;
png_uint_32 png_width, png_height;
int bit_depth, color_type, interlace_method, compression_method, filter_method;
png_get_IHDR(png_ptr, info_ptr, &png_width, &png_height, &bit_depth, &color_type,
&interlace_method, &compression_method, &filter_method);
image.m_width = png_width;
image.m_height = png_height;
// --- Transformations ---
// We want linear float RGB or RGBA output
// Handle palette -> RGB
if (color_type == PNG_COLOR_TYPE_PALETTE)
{
png_set_palette_to_rgb(png_ptr);
}
// Handle low bit depth grayscale -> 8 bit
if (color_type == PNG_COLOR_TYPE_GRAY && bit_depth < 8)
{
png_set_expand_gray_1_2_4_to_8(png_ptr);
bit_depth = 8; // Update bit depth after expansion
}
// Handle transparency chunk -> Alpha channel
if (png_get_valid(png_ptr, info_ptr, PNG_INFO_tRNS))
{
png_set_tRNS_to_alpha(png_ptr);
}
// Convert 16-bit -> 8-bit if needed (we handle 16 bit below, so maybe don't strip)
// if (bit_depth == 16) {
// png_set_strip_16(png_ptr);
// bit_depth = 8;
// }
// Convert grayscale -> RGB
if (color_type == PNG_COLOR_TYPE_GRAY || color_type == PNG_COLOR_TYPE_GRAY_ALPHA)
{
png_set_gray_to_rgb(png_ptr);
}
// Add alpha channel if missing but requested (we might always want RGBA internally)
// if (color_type == PNG_COLOR_TYPE_RGB || color_type == PNG_COLOR_TYPE_GRAY) {
// png_set_add_alpha(png_ptr, 0xFF, PNG_FILLER_AFTER); // Add opaque alpha
// }
// --- Gamma Handling ---
double file_gamma = 0.0;
bool is_srgb = (png_get_sRGB(png_ptr, info_ptr, nullptr) != 0);
if (is_srgb)
{
// If sRGB chunk is present, libpng can convert to linear for us
png_set_gamma(png_ptr, 1.0, 0.45455); // Request linear output (screen gamma 2.2)
image.m_isLinear = true;
image.m_colorSpaceName = "Linear sRGB";
}
else if (png_get_gAMA(png_ptr, info_ptr, &file_gamma))
{
// If gAMA chunk is present, convert to linear
png_set_gamma(png_ptr, 1.0, file_gamma);
image.m_isLinear = true;
image.m_colorSpaceName = "Linear Unknown (Gamma Corrected)";
}
else
{
// No gamma info, assume sRGB and convert manually later
image.m_isLinear = false; // Data read will be sRGB
image.m_colorSpaceName = "sRGB (Assumed)";
}
// Apply transformations
png_read_update_info(png_ptr, info_ptr);
// Get updated info after transformations
image.m_channels = png_get_channels(png_ptr, info_ptr);
bit_depth = png_get_bit_depth(png_ptr, info_ptr); // Update bit_depth after transforms
if (image.m_channels < 3)
{
LogWarning("libpng: Resulting image has < 3 channels after transforms. Handling as RGB.");
// Force RGB if needed? Be careful here. For simplicity, assume RGB/RGBA works.
}
// Allocate memory for the image data
size_t num_pixels = static_cast<size_t>(image.m_width) * image.m_height;
size_t total_floats = num_pixels * image.m_channels;
image.m_pixelData.resize(total_floats);
float *app_data_ptr = image.m_pixelData.data();
// Allocate row pointers
png_bytep *row_pointers = new png_bytep[image.m_height];
size_t row_bytes = png_get_rowbytes(png_ptr, info_ptr);
std::vector<unsigned char> image_buffer(row_bytes * image.m_height); // Read whole image at once
for (png_uint_32 i = 0; i < image.m_height; ++i)
{
row_pointers[i] = image_buffer.data() + i * row_bytes;
}
// Read the entire image
png_read_image(png_ptr, row_pointers);
// Convert the read data to linear float
unsigned char *buffer_ptr = image_buffer.data();
if (bit_depth == 8)
{
for (size_t i = 0; i < total_floats; ++i)
{
float val = static_cast<float>(buffer_ptr[i]) / 255.0f;
// Convert to linear if libpng didn't do it (i.e., no sRGB/gAMA chunk found)
app_data_ptr[i] = image.m_isLinear ? val : srgb_to_linear_approx(val);
}
}
else if (bit_depth == 16)
{
uint16_t *buffer_ptr16 = reinterpret_cast<uint16_t *>(buffer_ptr);
// PNG 16-bit uses network byte order (big-endian)
bool needs_swap = (png_get_uint_16((png_bytep) "\x01\x02") != 0x0102); // Check system endianness
for (size_t i = 0; i < total_floats; ++i)
{
uint16_t raw_val = buffer_ptr16[i];
if (needs_swap)
{ // Swap bytes if system is little-endian
raw_val = (raw_val >> 8) | (raw_val << 8);
}
float val = static_cast<float>(raw_val) / 65535.0f;
// Convert to linear if libpng didn't do it
app_data_ptr[i] = image.m_isLinear ? val : srgb_to_linear_approx(val);
}
}
else
{
LogError("libpng: Unsupported bit depth after transforms: " + std::to_string(bit_depth));
delete[] row_pointers;
png_destroy_read_struct(&png_ptr, &info_ptr, nullptr);
fclose(fp);
return std::nullopt;
}
// If we assumed sRGB and converted manually, update state
if (!image.m_isLinear)
{
image.m_isLinear = true;
image.m_colorSpaceName = "Linear sRGB (Assumed)";
}
// Clean up
delete[] row_pointers;
png_read_end(png_ptr, nullptr); // Finish reading remaining chunks
png_destroy_read_struct(&png_ptr, &info_ptr, nullptr);
fclose(fp);
// Note: PNG typically doesn't store EXIF in the same way as JPEG/TIFF.
// It can have text chunks (tEXt, zTXt, iTXt) which might hold metadata.
// Reading these requires additional libpng calls (png_get_text). Not implemented here.
return image;
}
// --- libtiff Loading ---
// Suppress libtiff warnings/errors (optional, can be noisy)
void tiffErrorHandler(const char *module, const char *fmt, va_list ap) { /* Do nothing */ }
void tiffWarningHandler(const char *module, const char *fmt, va_list ap) { /* Do nothing */ }
inline std::optional<AppImage> loadTiff(const std::string &filePath)
{
// Set custom handlers to suppress console output from libtiff
// TIFFSetErrorHandler(tiffErrorHandler);
// TIFFSetWarningHandler(tiffWarningHandler);
TIFF *tif = TIFFOpen(filePath.c_str(), "r");
if (!tif)
{
LogError("Cannot open TIFF file: " + filePath);
return std::nullopt;
}
AppImage image;
uint32_t w, h;
uint16_t bitsPerSample, samplesPerPixel, photometric, planarConfig;
TIFFGetFieldDefaulted(tif, TIFFTAG_IMAGEWIDTH, &w);
TIFFGetFieldDefaulted(tif, TIFFTAG_IMAGELENGTH, &h);
TIFFGetFieldDefaulted(tif, TIFFTAG_BITSPERSAMPLE, &bitsPerSample);
TIFFGetFieldDefaulted(tif, TIFFTAG_SAMPLESPERPIXEL, &samplesPerPixel);
TIFFGetFieldDefaulted(tif, TIFFTAG_PHOTOMETRIC, &photometric);
TIFFGetFieldDefaulted(tif, TIFFTAG_PLANARCONFIG, &planarConfig);
image.m_width = w;
image.m_height = h;
image.m_channels = samplesPerPixel; // Usually 1 (Gray) or 3 (RGB) or 4 (RGBA)
// --- Sanity Checks ---
if (w == 0 || h == 0 || samplesPerPixel == 0)
{
LogError("libtiff: Invalid dimensions or samples per pixel.");
TIFFClose(tif);
return std::nullopt;
}
if (bitsPerSample != 8 && bitsPerSample != 16 && bitsPerSample != 32)
{
// Note: 32-bit float TIFFs exist but require different handling
LogError("libtiff: Unsupported bits per sample: " + std::to_string(bitsPerSample) + ". Only 8/16 supported currently.");
TIFFClose(tif);
return std::nullopt;
}
if (photometric != PHOTOMETRIC_MINISBLACK && photometric != PHOTOMETRIC_MINISWHITE &&
photometric != PHOTOMETRIC_RGB && photometric != PHOTOMETRIC_PALETTE &&
photometric != PHOTOMETRIC_MASK && photometric != PHOTOMETRIC_SEPARATED /*CMYK?*/ &&
photometric != PHOTOMETRIC_LOGL && photometric != PHOTOMETRIC_LOGLUV)
{
LogWarning("libtiff: Unhandled photometric interpretation: " + std::to_string(photometric));
// We will try to read as RGB/Gray anyway... might be wrong.
}
// --- Data Reading ---
// Use TIFFReadRGBAImage for simplicity - converts many formats to RGBA uint32 internally
// Advantage: Handles various photometric interpretations, planar configs, palettes etc.
// Disadvantage: Always gives 8-bit RGBA, loses 16-bit precision. Less control.
// Alternative: Read scanlines manually (more complex, preserves bit depth)
// Let's try the manual scanline approach to preserve bit depth
size_t num_pixels = static_cast<size_t>(w) * h;
size_t total_values = num_pixels * samplesPerPixel; // Total uint8/uint16 values
image.m_pixelData.resize(total_values); // Resize for float output
image.m_isLinear = true; // Assume linear, correct later if gamma info found
image.m_colorSpaceName = "Linear Unknown (TIFF)"; // Default assumption
tmsize_t scanline_size = TIFFScanlineSize(tif);
std::vector<unsigned char> scanline_buffer(scanline_size);
float *app_data_ptr = image.m_pixelData.data();
float max_val = (bitsPerSample == 8) ? 255.0f : 65535.0f; // Normalization factor
if (planarConfig == PLANARCONFIG_CONTIG)
{
for (uint32_t row = 0; row < h; ++row)
{
if (TIFFReadScanline(tif, scanline_buffer.data(), row) < 0)
{
LogError("libtiff: Error reading scanline " + std::to_string(row));
TIFFClose(tif);
return std::nullopt;
}
// Process the contiguous scanline
if (bitsPerSample == 8)
{
unsigned char *buf_ptr = scanline_buffer.data();
for (size_t i = 0; i < w * samplesPerPixel; ++i)
{
*app_data_ptr++ = static_cast<float>(buf_ptr[i]) / max_val;
}
}
else
{ // bitsPerSample == 16
uint16_t *buf_ptr = reinterpret_cast<uint16_t *>(scanline_buffer.data());
for (size_t i = 0; i < w * samplesPerPixel; ++i)
{
*app_data_ptr++ = static_cast<float>(buf_ptr[i]) / max_val;
}
}
}
}
else if (planarConfig == PLANARCONFIG_SEPARATE)
{
// Read plane by plane - more complex, needs buffer per plane
LogWarning("libtiff: Planar configuration PLANARCONFIG_SEPARATE reading not fully implemented, data might be incorrect.");
// Basic attempt: Read all scanlines for each plane sequentially into the final buffer
size_t plane_stride = w * h;
for (uint16_t plane = 0; plane < samplesPerPixel; ++plane)
{
float *plane_start_ptr = image.m_pixelData.data() + plane; // Start at the channel offset
for (uint32_t row = 0; row < h; ++row)
{
if (TIFFReadScanline(tif, scanline_buffer.data(), row, plane) < 0)
{
LogError("libtiff: Error reading scanline " + std::to_string(row) + " plane " + std::to_string(plane));
TIFFClose(tif);
return std::nullopt;
}
// Process the separate scanline for this plane
if (bitsPerSample == 8)
{
unsigned char *buf_ptr = scanline_buffer.data();
float *current_pixel_in_plane = plane_start_ptr + row * w * samplesPerPixel;
for (uint32_t col = 0; col < w; ++col)
{
*current_pixel_in_plane = static_cast<float>(buf_ptr[col]) / max_val;
current_pixel_in_plane += samplesPerPixel; // Jump to next pixel's spot for this channel
}
}
else
{ // 16 bit
uint16_t *buf_ptr = reinterpret_cast<uint16_t *>(scanline_buffer.data());
float *current_pixel_in_plane = plane_start_ptr + row * w * samplesPerPixel;
for (uint32_t col = 0; col < w; ++col)
{
*current_pixel_in_plane = static_cast<float>(buf_ptr[col]) / max_val;
current_pixel_in_plane += samplesPerPixel;
}
}
}
}
}
else
{
LogError("libtiff: Unknown planar configuration: " + std::to_string(planarConfig));
TIFFClose(tif);
return std::nullopt;
}
// --- Post-processing based on Photometric interpretation ---
// Handle grayscale inversion
if (photometric == PHOTOMETRIC_MINISWHITE)
{
LogWarning("libtiff: Inverting MINISWHITE image.");
for (float &val : image.m_pixelData)
{
val = 1.0f - val; // Simple inversion
}
}
// TODO: Handle Palette -> RGB (needs reading the colormap tag)
if (photometric == PHOTOMETRIC_PALETTE)
{
LogWarning("libtiff: PHOTOMETRIC_PALETTE not fully handled. Image loaded as indexed.");
// Requires reading TIFFTAG_COLORMAP and expanding pixels
}
// TODO: Check for gamma tags or ICC profile tag
// uint16_t* icc_profile_count = nullptr;
// void* icc_profile_data = nullptr;
// if (TIFFGetField(tif, TIFFTAG_ICCPROFILE, &icc_profile_count, &icc_profile_data) && icc_profile_count && icc_profile_data) {
// image.m_iccProfile.resize(*icc_profile_count);
// std::memcpy(image.m_iccProfile.data(), icc_profile_data, *icc_profile_count);
// image.m_colorSpaceName = "Linear (Embedded ICC)"; // Or just "(Embedded ICC)"
// } else {
// // Check for gamma? Not standard. Assume sRGB/linear for now.
// }
// If no specific color info found, assume sRGB and convert to linear
// For TIFF, it's often safer to assume linear if 16-bit, sRGB if 8-bit without other info.
if (bitsPerSample == 8)
{
LogWarning("libtiff: Assuming 8-bit TIFF is sRGB. Converting to linear.");
for (float &val : image.m_pixelData)
{
val = srgb_to_linear_approx(val);
}
image.m_isLinear = true;
image.m_colorSpaceName = "Linear sRGB (Assumed)";
}
else
{
LogWarning("libtiff: Assuming 16-bit TIFF is already linear.");
image.m_isLinear = true;
image.m_colorSpaceName = "Linear Unknown (TIFF)";
}
TIFFClose(tif);
// Try loading EXIF using LibTiff directory reading or Exiv2 (not EasyExif)
// This basic example doesn't load EXIF from TIFFs.
// You could use Exiv2 here if integrated.
LogWarning("EXIF loading from TIFF not implemented in this example.");
return image;
}
} // namespace AppImageUtil
// --- AppImage Constructor Implementation ---
AppImage::AppImage(uint32_t width, uint32_t height, uint32_t channels)
: m_width(width), m_height(height), m_channels(channels), m_isLinear(true)
{
if (width > 0 && height > 0 && channels > 0)
{
try
{
m_pixelData.resize(static_cast<size_t>(width) * height * channels);
}
catch (const std::bad_alloc &e)
{
AppImageUtil::LogError("Failed to allocate memory for image: " + std::string(e.what()));
clear_image(); // Reset to empty state
throw; // Re-throw exception
}
}
// Default assumption is linear data in our internal format
m_colorSpaceName = "Linear Generic";
}
void AppImage::resize(uint32_t newWidth, uint32_t newHeight, uint32_t newChannels)
{
if (newChannels == 0)
newChannels = m_channels;
if (newChannels == 0)
newChannels = 3; // Default if was empty
m_width = newWidth;
m_height = newHeight;
m_channels = newChannels;
if (newWidth == 0 || newHeight == 0 || newChannels == 0)
{
m_pixelData.clear();
// Keep metadata? Optional.
}
else
{
try
{
m_pixelData.resize(static_cast<size_t>(newWidth) * newHeight * newChannels);
// Note: Resizing doesn't preserve pixel content intelligently.
// Consider adding different resize modes (clear, copy existing, etc.)
}
catch (const std::bad_alloc &e)
{
AppImageUtil::LogError("Failed to allocate memory during resize: " + std::string(e.what()));
clear_image();
throw;
}
}
}
void AppImage::clear_image()
{
m_width = 0;
m_height = 0;
m_channels = 0;
m_pixelData.clear();
m_metadata.clear();
m_iccProfile.clear();
m_colorSpaceName = "Unknown";
m_isLinear = true;
}
// --- loadImage Implementation ---
std::optional<AppImage> loadImage(const std::string &filePath)
{
using namespace AppImageUtil;
DetectedFileType type = detectFileType(filePath);
try
{
switch (type)
{
case DetectedFileType::RAW:
LogWarning("Detected type: RAW (using LibRaw)");
return loadRaw(filePath);
case DetectedFileType::JPEG:
LogWarning("Detected type: JPEG (using libjpeg)");
return loadJpeg(filePath);
case DetectedFileType::PNG:
LogWarning("Detected type: PNG (using libpng)");
return loadPng(filePath);
case DetectedFileType::TIFF:
LogWarning("Detected type: TIFF (using libtiff)");
// LibRaw can sometimes open TIFFs that contain RAW data. Try it first?
// For now, directly use libtiff.
return loadTiff(filePath);
case DetectedFileType::UNKNOWN:
default:
LogError("Unknown or unsupported file type: " + filePath);
return std::nullopt;
}
}
catch (const std::exception &e)
{
LogError("Exception caught during image loading: " + std::string(e.what()));
return std::nullopt;
}
catch (...)
{
LogError("Unknown exception caught during image loading.");
return std::nullopt;
}
}
// --- saveImage Implementation ---
namespace AppImageUtil
{
// --- libjpeg Saving ---
inline bool saveJpeg(const AppImage &image, const std::string &filePath, int quality)
{
if (image.getChannels() != 1 && image.getChannels() != 3)
{
LogError("libjpeg save: Can only save 1 (Grayscale) or 3 (RGB) channels. Image has " + std::to_string(image.getChannels()));
return false;
}
FILE *outfile = fopen(filePath.c_str(), "wb");
if (!outfile)
{
LogError("Cannot open file for JPEG writing: " + filePath);
return false;
}
jpeg_compress_struct cinfo;
JpegErrorManager jerr; // Use the same error manager as loading
// Setup error handling
cinfo.err = jpeg_std_error(&jerr.pub);
jerr.pub.error_exit = jpegErrorExit; // Use the same exit function
if (setjmp(jerr.setjmp_buffer))
{
// Error occurred during compression
jpeg_destroy_compress(&cinfo);
fclose(outfile);
return false;
}
// Initialize compression object
jpeg_create_compress(&cinfo);
jpeg_stdio_dest(&cinfo, outfile);
// Set parameters
cinfo.image_width = image.getWidth();
cinfo.image_height = image.getHeight();
cinfo.input_components = image.getChannels();
cinfo.in_color_space = (image.getChannels() == 1) ? JCS_GRAYSCALE : JCS_RGB;
jpeg_set_defaults(&cinfo);
jpeg_set_quality(&cinfo, std::max(1, std::min(100, quality)), TRUE /* limit to baseline-JPEG */);
// Could set density, comments, etc. here if needed using jpeg_set_... functions
// Start compressor
jpeg_start_compress(&cinfo, TRUE);
// Prepare 8-bit sRGB scanline buffer
int row_stride = cinfo.image_width * cinfo.input_components;
std::vector<unsigned char> scanline_buffer(row_stride);
const float *app_data = image.getData();
// Process scanlines
while (cinfo.next_scanline < cinfo.image_height)
{
unsigned char *buffer_ptr = scanline_buffer.data();
size_t row_start_index = static_cast<size_t>(cinfo.next_scanline) * cinfo.image_width * cinfo.input_components;
// Convert one row from linear float to 8-bit sRGB uchar
for (int i = 0; i < row_stride; ++i)
{
float linear_val = app_data[row_start_index + i];
float srgb_val = linear_to_srgb_approx(linear_val);
int int_val = static_cast<int>(std::round(srgb_val * 255.0f));
buffer_ptr[i] = static_cast<unsigned char>(std::max(0, std::min(255, int_val)));
}
JSAMPROW row_pointer[1];
row_pointer[0] = scanline_buffer.data();
jpeg_write_scanlines(&cinfo, row_pointer, 1);
}
// Finish compression and clean up
jpeg_finish_compress(&cinfo);
jpeg_destroy_compress(&cinfo);
fclose(outfile);
// --- Metadata Saving ---
LogWarning("JPEG EXIF/ICC Metadata saving is NOT implemented.");
// Saving metadata would typically involve:
// 1. Using Exiv2 library.
// 2. Opening the file *after* libjpeg saves the pixels.
// 3. Writing the metadata from image.m_metadata and image.m_iccProfile into the file structure.
return true;
}
// --- libpng Saving ---
inline bool savePng(const AppImage &image, const std::string &filePath, int bit_depth_out)
{
if (bit_depth_out != 8 && bit_depth_out != 16)
{
LogError("libpng save: Only 8 or 16 bit output supported.");
return false;
}
if (image.getChannels() < 1 || image.getChannels() > 4 || image.getChannels() == 2)
{
LogError("libpng save: Can only save 1 (Gray), 3 (RGB), or 4 (RGBA) channels. Image has " + std::to_string(image.getChannels()));
return false;
}
FILE *fp = fopen(filePath.c_str(), "wb");
if (!fp)
{
LogError("Cannot open file for PNG writing: " + filePath);
return false;
}
png_structp png_ptr = png_create_write_struct(PNG_LIBPNG_VER_STRING, nullptr, pngErrorFunc, pngWarningFunc);
if (!png_ptr)
{
LogError("libpng: png_create_write_struct failed");
fclose(fp);
return false;
}
png_infop info_ptr = png_create_info_struct(png_ptr);
if (!info_ptr)
{
LogError("libpng: png_create_info_struct failed");
png_destroy_write_struct(&png_ptr, nullptr);
fclose(fp);
return false;
}
// Setup jump buffer for error handling
jmp_buf jmpbuf;
if (setjmp(jmpbuf))
{
LogError("libpng: Error during write");
png_destroy_write_struct(&png_ptr, &info_ptr);
fclose(fp);
return false;
}
png_set_error_fn(png_ptr, reinterpret_cast<png_voidp>(&jmpbuf), pngErrorFunc, pngWarningFunc);
png_init_io(png_ptr, fp);
// Determine PNG color type
int color_type;
switch (image.getChannels())
{
case 1:
color_type = PNG_COLOR_TYPE_GRAY;
break;
case 3:
color_type = PNG_COLOR_TYPE_RGB;
break;
case 4:
color_type = PNG_COLOR_TYPE_RGB_ALPHA;
break;
default: /* Should have been caught earlier */
return false;
}
// Set IHDR chunk
png_set_IHDR(png_ptr, info_ptr, image.getWidth(), image.getHeight(),
bit_depth_out, color_type,
PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT);
// Set Gamma/sRGB info
bool save_as_srgb = (bit_depth_out == 8); // Convention: Save 8-bit as sRGB, 16-bit as linear
if (save_as_srgb)
{
png_set_sRGB_gAMA_and_cHRM(png_ptr, info_ptr, PNG_sRGB_INTENT_PERCEPTUAL);
LogWarning("libpng save: Saving 8-bit PNG with sRGB chunk.");
}
else
{ // 16-bit linear
png_set_gAMA(png_ptr, info_ptr, 1.0); // Explicitly linear gamma
LogWarning("libpng save: Saving 16-bit PNG with gamma 1.0 (linear).");
}
// Write header info
png_write_info(png_ptr, info_ptr);
// --- Prepare Data ---
std::vector<png_bytep> row_pointers(image.getHeight());
size_t values_per_row = static_cast<size_t>(image.getWidth()) * image.getChannels();
size_t bytes_per_value = (bit_depth_out == 8) ? 1 : 2;
size_t row_bytes = values_per_row * bytes_per_value;
std::vector<unsigned char> output_buffer(row_bytes * image.getHeight());
const float *app_data = image.getData();
bool needs_swap = (bit_depth_out == 16 && (png_get_uint_16((png_bytep) "\x01\x02") != 0x0102)); // Check endianness only for 16-bit
// Convert internal float data to target format row by row
for (uint32_t y = 0; y < image.getHeight(); ++y)
{
unsigned char *row_buf_ptr = output_buffer.data() + y * row_bytes;
row_pointers[y] = row_buf_ptr;
size_t row_start_index = static_cast<size_t>(y) * values_per_row;
if (bit_depth_out == 8)
{
unsigned char *uchar_ptr = row_buf_ptr;
for (size_t i = 0; i < values_per_row; ++i)
{
float linear_val = app_data[row_start_index + i];
float srgb_val = linear_to_srgb_approx(linear_val); // Convert to sRGB for 8-bit output
int int_val = static_cast<int>(std::round(srgb_val * 255.0f));
uchar_ptr[i] = static_cast<unsigned char>(std::max(0, std::min(255, int_val)));
}
}
else
{ // 16-bit
uint16_t *ushort_ptr = reinterpret_cast<uint16_t *>(row_buf_ptr);
for (size_t i = 0; i < values_per_row; ++i)
{
float linear_val = app_data[row_start_index + i];
// Clamp linear value before scaling for 16-bit output (0.0 to 1.0 range typical for linear PNG)
linear_val = std::fmax(0.0f, std::fmin(1.0f, linear_val));
int int_val = static_cast<int>(std::round(linear_val * 65535.0f));
uint16_t val16 = static_cast<uint16_t>(std::max(0, std::min(65535, int_val)));
if (needs_swap)
{ // Swap bytes for big-endian PNG format
ushort_ptr[i] = (val16 >> 8) | (val16 << 8);
}
else
{
ushort_ptr[i] = val16;
}
}
}
}
// Write image data
png_write_image(png_ptr, row_pointers.data());
// End writing
png_write_end(png_ptr, nullptr);
// Clean up
png_destroy_write_struct(&png_ptr, &info_ptr);
fclose(fp);
LogWarning("PNG Metadata saving (text chunks, ICC) is NOT implemented.");
return true;
}
// --- libtiff Saving ---
inline bool saveTiff(const AppImage &image, const std::string &filePath, int bit_depth_out)
{
if (bit_depth_out != 8 && bit_depth_out != 16)
{
LogError("libtiff save: Only 8 or 16 bit output supported.");
return false;
}
if (image.getChannels() < 1 || image.getChannels() > 4 || image.getChannels() == 2)
{
LogError("libtiff save: Can only save 1 (Gray), 3 (RGB), or 4 (RGBA) channels. Image has " + std::to_string(image.getChannels()));
return false;
}
TIFF *tif = TIFFOpen(filePath.c_str(), "w");
if (!tif)
{
LogError("Cannot open file for TIFF writing: " + filePath);
return false;
}
// --- Set Core TIFF Tags ---
TIFFSetField(tif, TIFFTAG_IMAGEWIDTH, image.getWidth());
TIFFSetField(tif, TIFFTAG_IMAGELENGTH, image.getHeight());
TIFFSetField(tif, TIFFTAG_SAMPLESPERPIXEL, static_cast<uint16_t>(image.getChannels()));
TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, static_cast<uint16_t>(bit_depth_out));
TIFFSetField(tif, TIFFTAG_ORIENTATION, ORIENTATION_TOPLEFT);
TIFFSetField(tif, TIFFTAG_PLANARCONFIG, PLANARCONFIG_CONTIG); // Interleaved is simpler
// Set Photometric Interpretation
uint16_t photometric;
if (image.getChannels() == 1)
{
photometric = PHOTOMETRIC_MINISBLACK; // Grayscale
}
else if (image.getChannels() >= 3)
{
photometric = PHOTOMETRIC_RGB; // RGB or RGBA
if (image.getChannels() == 4)
{
// Need to specify that the extra channel is Alpha
uint16_t extra_samples = 1;
uint16_t sample_info[] = {EXTRASAMPLE_ASSOCALPHA}; // Associated alpha
TIFFSetField(tif, TIFFTAG_EXTRASAMPLES, extra_samples, sample_info);
}
}
else
{
LogError("libtiff save: Unexpected channel count: " + std::to_string(image.getChannels()));
TIFFClose(tif);
return false;
}
TIFFSetField(tif, TIFFTAG_PHOTOMETRIC, photometric);
// Compression (optional, default is none)
TIFFSetField(tif, TIFFTAG_COMPRESSION, COMPRESSION_NONE);
// Examples: COMPRESSION_LZW, COMPRESSION_ADOBE_DEFLATE
// Rows per strip (can affect performance/compatibility)
// A sensible default is often related to scanline buffer size.
TIFFSetField(tif, TIFFTAG_ROWSPERSTRIP, TIFFDefaultStripSize(tif, (uint32_t)-1));
// Software Tag (optional)
TIFFSetField(tif, TIFFTAG_SOFTWARE, "AppImage Saver");
// --- Prepare and Write Data ---
size_t values_per_row = static_cast<size_t>(image.getWidth()) * image.getChannels();
size_t bytes_per_value = (bit_depth_out == 8) ? 1 : 2;
tmsize_t row_bytes = values_per_row * bytes_per_value;
std::vector<unsigned char> output_buffer(row_bytes); // Buffer for one row
const float *app_data = image.getData();
bool save_as_srgb = (bit_depth_out == 8); // Convention: 8-bit=sRGB, 16-bit=Linear
for (uint32_t y = 0; y < image.getHeight(); ++y)
{
unsigned char *row_buf_ptr = output_buffer.data();
size_t row_start_index = static_cast<size_t>(y) * values_per_row;
if (bit_depth_out == 8)
{
unsigned char *uchar_ptr = row_buf_ptr;
for (size_t i = 0; i < values_per_row; ++i)
{
float linear_val = app_data[row_start_index + i];
float srgb_val = linear_to_srgb_approx(linear_val); // Convert to sRGB
int int_val = static_cast<int>(std::round(srgb_val * 255.0f));
uchar_ptr[i] = static_cast<unsigned char>(std::max(0, std::min(255, int_val)));
}
}
else
{ // 16-bit
uint16_t *ushort_ptr = reinterpret_cast<uint16_t *>(row_buf_ptr);
for (size_t i = 0; i < values_per_row; ++i)
{
float linear_val = app_data[row_start_index + i];
// Clamp linear [0,1] before scaling
linear_val = std::fmax(0.0f, std::fmin(1.0f, linear_val));
int int_val = static_cast<int>(std::round(linear_val * 65535.0f));
ushort_ptr[i] = static_cast<uint16_t>(std::max(0, std::min(65535, int_val)));
// Note: TIFF uses native byte order by default, no swapping needed usually.
}
}
// Write the scanline
if (TIFFWriteScanline(tif, row_buf_ptr, y, 0) < 0)
{
LogError("libtiff save: Error writing scanline " + std::to_string(y));
TIFFClose(tif);
return false;
}
}
// Clean up
TIFFClose(tif);
LogWarning("TIFF EXIF/ICC Metadata saving is NOT implemented.");
// Saving metadata requires:
// 1. Using Exiv2 or LibTiff's directory writing functions *before* closing the file.
// 2. For ICC: TIFFSetField(tif, TIFFTAG_ICCPROFILE, count, data_ptr);
return true;
}
} // namespace AppImageUtil
namespace ImGuiImageViewerUtil {
// Linear float [0,1+] -> sRGB approx [0,1]
inline float linear_to_srgb_approx(float linearVal) {
if (linearVal <= 0.0f) return 0.0f;
linearVal = std::fmax(0.0f, std::fmin(1.0f, linearVal)); // Clamp for display
if (linearVal <= 0.0031308f) { return linearVal * 12.92f; }
else { return 1.055f * std::pow(linearVal, 1.0f / 2.4f) - 0.055f; }
}
// Round float to nearest integer
inline float Round(float f) { return ImFloor(f + 0.5f); }
} // namespace ImGuiImageViewerUtil
bool loadImageTexture(AppImage &appImage)
{
if (appImage.isEmpty() || appImage.getWidth() == 0 || appImage.getHeight() == 0)
{
AppImageUtil::LogError("loadImageTexture: Image is empty.");
return false;
}
if (!appImage.isLinear())
{
// This shouldn't happen if loadImage converts correctly, but good practice to check.
AppImageUtil::LogWarning("loadImageTexture: Warning - Image data is not linear. Pipeline expects linear input.");
// Ideally, convert to linear here if not already done. For now, proceed with caution.
}
const int width = static_cast<int>(appImage.getWidth());
const int height = static_cast<int>(appImage.getHeight());
const int channels = static_cast<int>(appImage.getChannels());
const float *linearData = appImage.getData();
size_t numFloats = static_cast<size_t>(width) * height * channels;
if (!linearData || numFloats == 0) {
AppImageUtil::LogError("loadImageTexture: Image data pointer is null or size is zero.");
return false;
}
// --- Determine OpenGL texture format ---
GLenum internalFormat;
GLenum dataFormat;
GLenum dataType = GL_FLOAT;
std::vector<float> textureDataBuffer; // Temporary buffer if we need to convert format (e.g., RGB -> RGBA)
const float* dataPtr = linearData;
if (channels == 1) {
internalFormat = GL_R16F; // Single channel, 16-bit float
dataFormat = GL_RED;
// Expand Grayscale to RGBA for easier shader handling (optional, shaders could handle GL_RED)
// Example: Expand to RGBA float buffer
textureDataBuffer.resize(static_cast<size_t>(width) * height * 4);
float* outPtr = textureDataBuffer.data();
for(int i = 0; i < width * height; ++i) {
float val = linearData[i];
*outPtr++ = val;
*outPtr++ = val;
*outPtr++ = val;
*outPtr++ = 1.0f; // Alpha
}
internalFormat = GL_RGBA16F; // Use RGBA16F if expanding
dataFormat = GL_RGBA;
dataPtr = textureDataBuffer.data(); // Point to the new buffer
AppImageUtil::LogWarning("loadImageTexture: Expanding 1-channel to RGBA16F for texture.");
} else if (channels == 3) {
internalFormat = GL_RGBA16F; // Store as RGBA, easier for FBOs/blending
dataFormat = GL_RGBA;
// Need to convert RGB float -> RGBA float
textureDataBuffer.resize(static_cast<size_t>(width) * height * 4);
float* outPtr = textureDataBuffer.data();
const float* inPtr = linearData;
for(int i = 0; i < width * height; ++i) {
*outPtr++ = *inPtr++; // R
*outPtr++ = *inPtr++; // G
*outPtr++ = *inPtr++; // B
*outPtr++ = 1.0f; // A
}
dataPtr = textureDataBuffer.data(); // Point to the new buffer
AppImageUtil::LogWarning("loadImageTexture: Expanding 3-channel RGB to RGBA16F for texture.");
} else if (channels == 4) {
internalFormat = GL_RGBA16F; // Native RGBA
dataFormat = GL_RGBA;
dataPtr = linearData; // Use original data directly
} else {
AppImageUtil::LogError("loadImageTexture: Unsupported number of channels: " + std::to_string(channels));
return false;
}
// --- Upload to OpenGL Texture ---
GLint lastTexture;
glGetIntegerv(GL_TEXTURE_BINDING_2D, &lastTexture);
if (appImage.m_textureId == 0) {
glGenTextures(1, &appImage.m_textureId);
AppImageUtil::LogWarning("loadImageTexture: Generated new texture ID: " + std::to_string(appImage.m_textureId));
} else {
AppImageUtil::LogWarning("loadImageTexture: Reusing texture ID: " + std::to_string(appImage.m_textureId));
}
glBindTexture(GL_TEXTURE_2D, appImage.m_textureId);
// Use GL_LINEAR for smoother results when zooming/scaling in the viewer, even if processing is nearest neighbor.
// The processing pipeline itself uses FBOs, textures don't need mipmaps typically.
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glPixelStorei(GL_UNPACK_ALIGNMENT, 1); // Ensure correct alignment, especially for RGB data
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); // Data is contiguous
// Check if texture dimensions/format need updating
bool needsTexImage = true;
if (appImage.m_textureWidth == width && appImage.m_textureHeight == height) {
// Could potentially use glTexSubImage2D if format matches, but glTexImage2D is safer
// if the internal format might change or if it's the first load.
// For simplicity, we'll just recreate with glTexImage2D.
AppImageUtil::LogWarning("loadImageTexture: Texture dimensions match, overwriting with glTexImage2D.");
} else {
AppImageUtil::LogWarning("loadImageTexture: Texture dimensions or format mismatch, recreating with glTexImage2D.");
}
glTexImage2D(GL_TEXTURE_2D, 0, internalFormat, width, height, 0, dataFormat, dataType, dataPtr);
GLenum err = glGetError();
if (err != GL_NO_ERROR) {
AppImageUtil::LogError("loadImageTexture: OpenGL Error after glTexImage2D: " + std::to_string(err));
glBindTexture(GL_TEXTURE_2D, lastTexture); // Restore previous binding
// Consider deleting the texture ID if creation failed badly?
if (appImage.m_textureId != 0) {
glDeleteTextures(1, &appImage.m_textureId);
appImage.m_textureId = 0;
}
return false;
} else {
AppImageUtil::LogWarning("loadImageTexture: glTexImage2D successful.");
}
// Optional: Generate mipmaps if you want smoother downscaling *in the final view*
// glGenerateMipmap(GL_TEXTURE_2D);
// glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
glBindTexture(GL_TEXTURE_2D, lastTexture); // Restore previous binding
appImage.m_textureWidth = width;
appImage.m_textureHeight = height;
AppImageUtil::LogWarning("loadImageTexture: Successfully loaded linear data into texture ID " + std::to_string(appImage.m_textureId));
return true;
}
// --- saveImage Implementation ---
bool saveImage(const AppImage &image,
const std::string &filePath,
ImageSaveFormat format,
int quality)
{
using namespace AppImageUtil;
if (image.isEmpty())
{
LogError("Cannot save an empty image.");
return false;
}
// Ensure internal data is linear before saving (or handle conversion if needed)
if (!image.isLinear())
{
LogWarning("Attempting to save non-linear internal data. Results may be incorrect if conversion to target space isn't handled properly.");
// Ideally, convert to linear here if required by the saving functions.
// For this implementation, we assume the saving functions expect linear input
// and perform the linear -> target space conversion (e.g., linear -> sRGB).
}
try
{
switch (format)
{
case ImageSaveFormat::JPEG:
return saveJpeg(image, filePath, quality);
case ImageSaveFormat::PNG_8:
return savePng(image, filePath, 8);
case ImageSaveFormat::PNG_16:
return savePng(image, filePath, 16);
case ImageSaveFormat::TIFF_8:
return saveTiff(image, filePath, 8);
case ImageSaveFormat::TIFF_16:
return saveTiff(image, filePath, 16);
case ImageSaveFormat::UNKNOWN:
default:
LogError("Unknown or unsupported save format specified.");
return false;
}
}
catch (const std::exception &e)
{
LogError("Exception caught during image saving: " + std::string(e.what()));
return false;
}
catch (...)
{
LogError("Unknown exception caught during image saving.");
return false;
}
}
#endif // APP_IMAGE_IMPLEMENTATION
#endif // APP_IMAGE_H