gowebp

README

gowebp

Lossless WebP (VP8L) encoder in pure Go with no external runtime dependencies.

Status

• Encoder supports lossless VP8L only (no animation).
• Transform pipeline includes predictor, color transform, subtract green, and color cache.
• Near-lossless quantization is implemented for RGB (alpha left untouched).
• Optional meta Huffman groups are supported via HuffmanGroupBits.

Current Limitations

• Transform image selection is fixed (no adaptive block analysis).
• LZ77 and cache heuristics are basic; compression is functional but not tuned.
• Huffman groups use per-tile histograms with simple deduplication (no rate optimization).
• Color cache is supported only in the main image stream (not transform images).
• No streaming API; encoder builds full bitstream in memory.

Usage

go get github.com/lEx0/gowebp

opts := webp.AutoOptions(img)
if err := webp.Encode(&buf, img, &opts); err != nil {
    // handle error
}

if err := webp.Encode(&buf, img, &webp.Options{
	NearLossless: 50,
	ColorCacheBits: 4,
}); err != nil {
    // handle error
}

LZ77 Options

• LZ77Mode selects the match-finding strategy:
  • LZ77ModeAuto (default): current fast hashed matcher.
  • LZ77ModeNaive: simple O(n*window) matcher (lower memory, slower).
  • LZ77ModeHash: hashed ring buffer (faster, higher memory).
• LZ77HashBits controls hash table size ([8..16]), default 11.
• LZ77Chain controls candidates per bucket (power of two, [1..8]), default 2.

Validation

• Integration tests compare pixel output via golang.org/x/image/webp and dwebp when available.

Benchmarks

goos: darwin
goarch: arm64
pkg: github.com/lEx0/gowebp/webp
cpu: Apple M1 Pro

BenchmarkEncodeSolid128
BenchmarkEncodeSolid128-8            	    6476	    161514 ns/op	  793248 B/op	      99 allocs/op
BenchmarkEncodeSolid128WithCache
BenchmarkEncodeSolid128WithCache-8   	    6489	    173089 ns/op	  793983 B/op	      95 allocs/op
BenchmarkEncodeGradient256
BenchmarkEncodeGradient256-8         	   10000	    106664 ns/op	  129658 B/op	    1011 allocs/op
BenchmarkEncodeNoise64
BenchmarkEncodeNoise64-8             	   20358	     58251 ns/op	  231128 B/op	     269 allocs/op
BenchmarkEncodeAlpha64
BenchmarkEncodeAlpha64-8             	    9765	    113484 ns/op	  251600 B/op	     640 allocs/op
BenchmarkEncodeSolid512
BenchmarkEncodeSolid512-8            	     704	   1655700 ns/op	12117409 B/op	      97 allocs/op

Near-Lossless

• NearLossless is a coarse quantization level in the range [0..100] (0 disables).
• Current step mapping is intentionally simple and may change as tuning improves.
• ColorCacheBits must be in [0..11] (0 disables).
• HuffmanGroupBits must be 0 or in [2..9].

Roadmap

• Rate-optimized meta Huffman groups.
• Adaptive transform images (predictor/color transform tuning).
• Streaming encoder API.