image/color: YcCbCr conversion to RGBA bug

[qeedquan]

The image attached is decoded to color.YCbCr, if you do image encoding directly on it, ie, png.Encode(out, img) then it will generate the correct output, however, if you loop through it and use RGBA() to build the image, it will give you a garbled output as shown below.

The regression happened after this commit:

Author: Nigel Tao <nigeltao@golang.org>  2015-03-25 18:47:24
Committer: Nigel Tao <nigeltao@golang.org>  2015-03-26 18:30:55
Parent: b5c3a9e572a1257c0db47d74b45f8e03f2f91f27 (image: add image.CMYK and color.CMYK types.)
Child:  0def13ac3f5a5e9f8e5540d3f5469cd469bddfad (image/color: have CMYK.RGBA work in 16-bit color, per the Color interface.)
Branches: master, remotes/origin/dev.ssa, remotes/origin/master
Follows: go1.4beta1
Precedes: go1.5beta1

    image/color: have YCbCr.RGBA work in 16-bit color, per the Color
    interface.

    Change-Id: Ie025753df08ae93e7a5095a3426aff15fa2016fd
    Reviewed-on: https://go-review.googlesource.com/8073
    Reviewed-by: Rob Pike <r@golang.org>

I tried reverting the RGBA() code for YCbCr code to the original one and that works fine

package main

import (
    "image"
    "image/color"
    _ "image/jpeg"
    "image/png"
    "log"
    "os"
)

func main() {
    r, err := os.Open("start_screen.jpg")
    check(err)
    defer r.Close()

    img, _, err := image.Decode(r)
    check(err)

    b := img.Bounds()
    rgba := image.NewRGBA(b)
    for y := b.Min.Y; y < b.Max.Y; y++ {
        for x := b.Min.X; x < b.Max.X; x++ {
            r32, g32, b32, a32 := img.At(x, y).RGBA()
            c := color.RGBA{uint8(r32), uint8(g32), uint8(b32), uint8(a32)}
            rgba.SetRGBA(x, y, c)
        }
    }

    w, err := os.Create("test.png")
    check(err)
    defer w.Close()

    err = png.Encode(w, rgba)
    check(err)
}

func check(err error) {
    if err != nil {
        log.Fatal(err)
    }
}

[nigeltao]

The bug is here:

r32, g32, b32, a32 := img.At(x, y).RGBA()
c := color.RGBA{uint8(r32), uint8(g32), uint8(b32), uint8(a32)}

http://golang.org/pkg/image/color/#Color says that "each value [returned by RGBA] ranges within [0, 0xFFFF]". Thus, the r32 value is effectively a 16-bit value (as a uint32).

You then say "uint8(r32)", which takes the 8 low bits, not the 8 high bits, of that 16-bit value. Instead, it should be "uint8(r32 >> 8)".

Separately, if you want to convert a YCbCr image to an RGBA image, you don't have to roll your own loop. Instead, the easiest way to do that is the "Converting an Image to RGBA" section of http://blog.golang.org/go-imagedraw-package

[qeedquan]

This was for a C pixel buffer, I ended up using color.RGBAModel.Convert().(color.RGBA), thanks for noticing the bug in the code.

[nigeltao]

This is becoming tangential, but I don't think the C-ness of the pixel buffer makes a difference.

Separately, using color.RGBAModel is fine, but note that it pretty much just does what I recommended to do. The rgbaModel function in http://golang.org/src/image/color/color.go says:

r, g, b, a := c.RGBA()
return RGBA{uint8(r >> 8), uint8(g >> 8), uint8(b >> 8), uint8(a >> 8)}