# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Project Overview

Chrome is a high-performance C++17 port of negfix8, a tool for processing scanned color negative film. It inverts and color-corrects 16-bit linear Rec.709 TIFF images from film scanners.

## Build Commands

```bash
# Configure and build
cmake -B build/ -S .
cd build && make

# The executable is built at build/chrome
```

**Dependencies:** OpenCV (core, imgproc, imgcodecs), OpenMP (optional, for parallelization)

## Testing

Manual testing only. Compare output against reference files in `test_data/`:
- `test_data/test_input.tiff` - Sample input scan
- `test_data/test_expected.tiff` - Reference output from original negfix8
- `test_data/current_output.tiff` - Current implementation output

```bash
./build/chrome -i test_data/test_input.tiff -o test_data/current_output.tiff
# Then visually compare or diff against test_data/test_expected.tiff
```

## Usage

```bash
./chrome -i input.tif [-o output.tif] [flags]

Flags:
  -g <val>    Set gamma (default 2.15)
  -cs         Enable contrast stretching
  -sat        Enable saturation boost (+20%)
  -m          Mirror image horizontally
  -r <N>      Binning (scale down by 1/N)
  -c <file>   Create profile and save to file (exits after)
  -u <file>   Use existing profile
```

## Architecture

Single-file implementation in `main.cpp` with this pipeline:

1. **Settings/Profile structs** - CLI options and image analysis data
2. **Image analysis** (`analyzeImage`) - Finds min/max per channel after shave+blur preprocessing
3. **Profile save/load** - Stores derived gamma corrections and offset values
4. **LUT generation** (`generateLUTs`) - Creates 65536-entry lookup tables for each channel using OpenMP
5. **Pixel processing** - Applies LUTs in parallel via OpenMP
6. **Post-processing** - Optional saturation boost, contrast stretch, mirror
7. **Output** - Writes LZW-compressed TIFF

The core algorithm inverts negative film by dividing film base values by pixel values, applies per-channel gamma correction to neutralize orange mask, then applies global gamma.