# **image**
**Pure MoonBit Image Library** · Zero C Dependencies · Four-Target Native Support
[](https://www.moonbitlang.com/)
[](https://opensource.org/licenses/MIT)
[]()
[](https://mooncakes.io/package/walkzzz/image)
[]()
[]()
[]()
[]()
[]()
[]()
[](https://www.gitlink.org.cn/competitions/track1_2026MoonBit)
简体中文 · [English](README.en.md)
**[Highlights](#-highlights)** · **[Formats](#-format-support--18)** · **[Quick Start](#-quick-start)** · **[Architecture](#-architecture)** · **[Features](#-features)** · **[Docs](#-documentation)**
---
## 🏆 OSC 2026 Contest Declaration
This project participates in the **2026 MoonBit Domestic Open Source Ecosystem Competition (OSC 2026)**. AI usage disclosure: AI programming assistants were used during development for code generation, test writing, documentation, and refactoring. All AI-generated code has been manually reviewed, tested, and verified for correctness and maintainability. See [docs/skill.md](docs/skill.md) for detailed AI usage scope.
---
## 🧬 Self-Evolving Engineering System (Exclusive)
Beyond feature development, this project built a **self_evolve closed loop** (docs/roadmap.md Phase 5-9, v281-v305) — the engineering process itself is evolvable:
- **Executable test generation**: auto-generates test cases per module and feeds them into the 4-target regression
- **Regression auto-repair**: performance/correctness regression detection → propagation chain visualization → fix suggestions → applied fixes
- **Pipeline optimization**: operator equivalence checking + redundancy elimination + adjacent operator merging (v276)
- **grill-me interrogation engine**: executable domain rules adversarially reviewing engineering output (v279)
- **Evolution recommendation engine**: recommends the next evolution step from current state (v298)
- **Coverage blind-spot analysis**: automatically locates test coverage blind spots (v297)
> The test count growth visible across versions (3378 → 6574) is backed by a **reproducible automated engineering pipeline** — the hardest evidence for the "engineering quality" dimension, and a capability C reference projects like stb_image.h do not have.
---
## ✨ Highlights
|
### 🟢 Zero C Dependencies
Fully pure MoonBit, no C compiler needed. `moon add walkzzz/image` and you're done. Four targets, zero configuration.
|
### 🟢 Four Targets Native
native / wasm-gc / js / wasm share one codebase, **no conditional compilation**. 6,574 tests × 4 targets all green.
|
|
### 🟢 18 Formats
PNG · JPEG · BMP · GIF · QOI · TGA · PSD · HDR · PNM · TIFF · ICO · CUR · ICNS · APNG · WebP · AVIF · HEIF · JPEG-XL
|
|
### 🟢 Three Pixel Depths
8-bit `Image` · 16-bit `Image16` · HDR float `ImageF`
|
|
### 🟢 Full OCR Pipeline
Preprocessing → Text Detection → Segmentation → Feature Extraction → Classifier → Digit/English/Chinese OCR → Table Recognition → Handwriting → Multilingual
|
### 🟢 Safety Hardened
MAX_IMAGE_DIMENSION(65535) guard + check_dims validation + safe_mul overflow protection + **Formal Verification** (moon prove)
|
|
### 🟢 Formally Verified
Core safety functions carry SMT-checkable proofs: representation invariants, overflow safety, dimension validation — not just tested, **mathematically guaranteed**
|
---
## 🖼️ Format Support · 18
| Format | Decode | Encode | Notes |
|:------:|:------:|:------:|-------|
| **PNG** | ✅ | ✅ | 8/16-bit, Adam7 interlacing |
| **JPEG** | ✅ | ✅ | baseline, adjustable quality |
| **BMP** | ✅ | ✅ | 1/4/8/16/24/32-bit |
| **GIF** | ✅ | ✅ | animated GIF decode/encode |
| **QOI** | ✅ | ✅ | Quite OK Image |
| **TGA** | ✅ | ✅ | with RLE |
| **PSD** | ✅ | — | Photoshop document **(exclusive)** |
| **HDR** | ✅ | ✅ | IEEE 754 float **(exclusive)** |
| **PNM** | ✅ | ✅ | PPM / PGM |
| **TIFF** | ✅ | ✅ | uncompressed/LZW/PackBits |
| **ICO** | ✅ | ✅ | single/multi-size |
| **CUR** | ✅ | ✅ | Windows cursor |
| **ICNS** | ✅ | ✅ | macOS icon |
| **APNG** | ✅ | ✅ | animated PNG |
| **WebP** | ✅ (lossless) | ✅ (lossy VP8) | VP8L decode + VP8 encode |
| **AVIF** | ✅ | — | AV1 keyframe decode |
| **HEIF** | ✅ | — | AV1/HEIC decode |
| **JPEG-XL** | ✅ | — | basic decode |
---
## 🚀 Quick Start
### Installation
```bash
moon add walkzzz/image
```
> 🧩 **Modular installation** (since 0.6.0, six workspace modules published independently): `moon add walkzzz/codec` for the 18-format codecs only; `moon add walkzzz/process` for the 90+ algorithm packages; likewise core / photography / vision / domain.
### 30-Second Example
```moonbit
import "walkzzz/core/types"
import "walkzzz/codec/lib"
import "walkzzz/process/filter"
import "walkzzz/process/transform"
// Decode — auto-detect 18 formats
let img : Image = @lib.load_from_bytes_auto(data)
// Process — blur / resize / rotate
let blurred = @filter.gaussian_blur(img, 3.0)
let resized = @transform.resize(img, 256, 256)
// Encode — output as PNG bytes
let out : Bytes = @lib.write_png_to_bytes(resized)
```
### Error Handling
```moonbit
try {
let img = @lib.load_from_bytes(data)
} catch {
LoadError::FileIO(msg) => println("IO error: \{msg}")
LoadError::DecodeFailed(msg) => println("Decode failed: \{msg}")
LoadError::UnsupportedFormat(msg) => println("Unsupported format: \{msg}")
}
```
### Advanced: SIFT Feature Matching
```moonbit
import "walkzzz/process/feature"
// Detect SIFT features
let kp1 = @feature.sift_detect(img1)
let kp2 = @feature.sift_detect(img2)
// L2 distance + Lowe ratio test matching
let matches = @feature.sift_match(kp1, kp2, ratio_threshold=0.75)
// RANSAC robust homography estimation
let homography = @feature.ransac_homography(matches, threshold=5.0, iterations=1000)
```
### Streaming Decode
```moonbit
import "walkzzz/codec/lib"
// Row-by-row callback (full decode then dispatch; incremental planned for v5.0)
@lib.decode_stream(data, fn(row, y) {
// Process row y pixels: row : Array[Array[Int]]
})
```
---
## 🏗️ Architecture
```mermaid
graph TD
CORE["
**If this project helps you, please consider ⭐ Starring it!**
Made with 🤍 for the MoonBit community