Files
geo/server/internal/shared/digitocr/digitocr.go
T
root 9d6181260a
Desktop Client Build / Resolve Build Metadata (push) Successful in 43s
Frontend CI / Frontend (push) Successful in 3m49s
Backend CI / Backend (push) Failing after 7m10s
Desktop Client Build / Build Desktop Client (push) Successful in 23m4s
Desktop Client Build / Publish Client Artifacts to NAS (push) Successful in 28s
feat(media-supply): add media resource supply marketplace
Introduce an end-to-end media-supply feature: tenant-side resource sync
service/worker backed by a Meijiequan supplier client, ops-side management
APIs, and admin/ops web views for resources, orders, favorites and
submission. Adds a shared digitocr helper, MediaSupply config blocks for
tenant and ops, shared types, and migrations for supplier media resources,
price overrides, customer visibility and order refunds.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-05-29 23:17:01 +08:00

359 lines
8.2 KiB
Go

// Package digitocr 识别固定字体、干净背景的多位数字图(如 4 位橙色数码字验证码)。
// 思路:橙色阈值二值化 → 找前景 bbox → 等分 N 列 → 每块归一到固定网格 → 与 0-9 模板做 Hamming 距离。
package digitocr
import (
"errors"
"fmt"
"image"
"image/color"
_ "image/png"
"io"
"os"
"sort"
)
const (
GridW = 10 // 单数字画布宽(覆盖最宽字符 + 余量)
GridH = 18 // 单数字画布高
)
// Bitmap 单个数字归一后的位图(按行展开)。
type Bitmap [GridW * GridH]uint8
// Options 控制识别行为。零值即默认 4 位、橙色前景。
type Options struct {
Digits int // 期望位数,默认 4
IsForeground func(color.Color) bool // 自定义前景判定,nil 时用 IsOrange
}
// Recognize 从文件路径读图并识别。
func Recognize(path string, opts Options) (string, error) {
f, err := os.Open(path)
if err != nil {
return "", err
}
defer f.Close()
return RecognizeReader(f, opts)
}
// RecognizeReader 从 io.Reader 读图并识别。
func RecognizeReader(r io.Reader, opts Options) (string, error) {
img, _, err := image.Decode(r)
if err != nil {
return "", err
}
return RecognizeImage(img, opts)
}
// RecognizeImage 对已解码的 image.Image 做识别。
func RecognizeImage(img image.Image, opts Options) (string, error) {
if opts.Digits == 0 {
opts.Digits = 4
}
if opts.IsForeground == nil {
opts.IsForeground = IsOrange
}
bin, bbox := Binarize(img, opts.IsForeground)
if bbox.Empty() {
return "", errors.New("digitocr: no foreground pixels")
}
cells := SegmentDigits(bin, bbox, opts.Digits)
out := make([]byte, len(cells))
for i, cell := range cells {
bm := Normalize(bin, cell)
out[i] = '0' + byte(MatchDigit(bm))
}
if len(out) != opts.Digits {
return string(out), fmt.Errorf("digitocr: expected %d digits, segmented %d", opts.Digits, len(out))
}
return string(out), nil
}
// SegmentDigits 优先用 8 连通分量切分,每个数字应是一个 CC。
// CC 数不匹配时退回列投影找零列;再不匹配退回等宽切分。
func SegmentDigits(bin [][]bool, bbox image.Rectangle, want int) []image.Rectangle {
if rects := ConnectedComponents(bin); len(rects) == want {
return rects
}
if rects := projectionSplit(bin, bbox); len(rects) == want {
return rects
}
// 兜底:等宽切分
out := make([]image.Rectangle, want)
cellW := bbox.Dx() / want
for i := 0; i < want; i++ {
out[i] = image.Rect(
bbox.Min.X+i*cellW, bbox.Min.Y,
bbox.Min.X+(i+1)*cellW, bbox.Max.Y,
)
if i == want-1 {
out[i].Max.X = bbox.Max.X
}
}
return out
}
// ConnectedComponents 返回所有 8 连通前景分量的 bbox,按左上 x 升序排列。
// 小于 minPixels 的噪点分量会被丢弃(默认 2)。
func ConnectedComponents(bin [][]bool) []image.Rectangle {
const minPixels = 2
h := len(bin)
if h == 0 {
return nil
}
w := len(bin[0])
visited := make([][]bool, h)
for i := range visited {
visited[i] = make([]bool, w)
}
var rects []image.Rectangle
stack := make([][2]int, 0, 64)
for y := 0; y < h; y++ {
for x := 0; x < w; x++ {
if !bin[y][x] || visited[y][x] {
continue
}
minX, minY := x, y
maxX, maxY := x, y
pixCount := 0
stack = append(stack[:0], [2]int{x, y})
visited[y][x] = true
for len(stack) > 0 {
p := stack[len(stack)-1]
stack = stack[:len(stack)-1]
pixCount++
if p[0] < minX {
minX = p[0]
}
if p[1] < minY {
minY = p[1]
}
if p[0] > maxX {
maxX = p[0]
}
if p[1] > maxY {
maxY = p[1]
}
for dy := -1; dy <= 1; dy++ {
for dx := -1; dx <= 1; dx++ {
if dx == 0 && dy == 0 {
continue
}
nx, ny := p[0]+dx, p[1]+dy
if nx >= 0 && nx < w && ny >= 0 && ny < h && bin[ny][nx] && !visited[ny][nx] {
visited[ny][nx] = true
stack = append(stack, [2]int{nx, ny})
}
}
}
}
if pixCount >= minPixels {
rects = append(rects, image.Rect(minX, minY, maxX+1, maxY+1))
}
}
}
sort.Slice(rects, func(i, j int) bool { return rects[i].Min.X < rects[j].Min.X })
return rects
}
func projectionSplit(bin [][]bool, bbox image.Rectangle) []image.Rectangle {
cols := make([]int, bbox.Dx())
for y := bbox.Min.Y; y < bbox.Max.Y; y++ {
for x := bbox.Min.X; x < bbox.Max.X; x++ {
if bin[y][x] {
cols[x-bbox.Min.X]++
}
}
}
var rects []image.Rectangle
inRun, runStart := false, 0
for i := 0; i <= len(cols); i++ {
present := i < len(cols) && cols[i] > 0
if present && !inRun {
inRun = true
runStart = i
} else if !present && inRun {
inRun = false
rects = append(rects, image.Rect(
bbox.Min.X+runStart, bbox.Min.Y,
bbox.Min.X+i, bbox.Max.Y,
))
}
}
return rects
}
// IsOrange 经验阈值:识别图中那种橙色前景像素。
// 如背景色调差异较大,可换 HSV 判 H∈[10°,30°]。
func IsOrange(c color.Color) bool {
r, g, b, _ := c.RGBA()
r8, g8, b8 := r>>8, g>>8, b>>8
return r8 > 200 && g8 > 80 && g8 < 180 && b8 < 100
}
// Binarize 把图像转成 bool 矩阵 + 所有前景像素的边界框。
// 默认会通过 StripFrame 去掉与图像边缘相连的"外框"前景(如圆角矩形装饰边)。
func Binarize(img image.Image, fg func(color.Color) bool) ([][]bool, image.Rectangle) {
b := img.Bounds()
w, h := b.Dx(), b.Dy()
bin := make([][]bool, h)
for y := 0; y < h; y++ {
bin[y] = make([]bool, w)
for x := 0; x < w; x++ {
if fg(img.At(b.Min.X+x, b.Min.Y+y)) {
bin[y][x] = true
}
}
}
StripFrame(bin)
return bin, computeBBox(bin)
}
// StripFrame 用 8 邻接 flood fill 从图像四边把任何相连的前景"吃掉"。
// 适用于验证码常见的圆角矩形装饰边——只要数字本身和外框不相连,外框就会被清干净。
func StripFrame(bin [][]bool) {
h := len(bin)
if h == 0 {
return
}
w := len(bin[0])
type pt struct{ x, y int }
var stack []pt
push := func(x, y int) {
if x >= 0 && x < w && y >= 0 && y < h && bin[y][x] {
bin[y][x] = false
stack = append(stack, pt{x, y})
}
}
for x := 0; x < w; x++ {
push(x, 0)
push(x, h-1)
}
for y := 0; y < h; y++ {
push(0, y)
push(w-1, y)
}
for len(stack) > 0 {
p := stack[len(stack)-1]
stack = stack[:len(stack)-1]
for dy := -1; dy <= 1; dy++ {
for dx := -1; dx <= 1; dx++ {
if dx == 0 && dy == 0 {
continue
}
push(p.x+dx, p.y+dy)
}
}
}
}
func computeBBox(bin [][]bool) image.Rectangle {
h := len(bin)
if h == 0 {
return image.Rectangle{}
}
w := len(bin[0])
minX, minY := w, h
maxX, maxY := -1, -1
for y := 0; y < h; y++ {
for x := 0; x < w; x++ {
if bin[y][x] {
if x < minX {
minX = x
}
if y < minY {
minY = y
}
if x > maxX {
maxX = x
}
if y > maxY {
maxY = y
}
}
}
}
if maxX < 0 {
return image.Rectangle{}
}
return image.Rect(minX, minY, maxX+1, maxY+1)
}
// Normalize 把 cell 内的前景按原尺寸贴到 GridW x GridH 画布的左上角。
// 字体固定时不做缩放采样,模板就是真实像素,区分度最高。
// 若数字尺寸超出画布,多出的右/下部分被截断(GridW/GridH 应当设大于最大字符)。
func Normalize(bin [][]bool, cell image.Rectangle) Bitmap {
minX, minY := cell.Max.X, cell.Max.Y
maxX, maxY := cell.Min.X-1, cell.Min.Y-1
for y := cell.Min.Y; y < cell.Max.Y; y++ {
if y < 0 || y >= len(bin) {
continue
}
for x := cell.Min.X; x < cell.Max.X; x++ {
if x < 0 || x >= len(bin[y]) {
continue
}
if bin[y][x] {
if x < minX {
minX = x
}
if y < minY {
minY = y
}
if x > maxX {
maxX = x
}
if y > maxY {
maxY = y
}
}
}
}
var bm Bitmap
if minX > maxX {
return bm
}
for y := minY; y <= maxY; y++ {
gy := y - minY
if gy >= GridH {
break
}
for x := minX; x <= maxX; x++ {
gx := x - minX
if gx >= GridW {
break
}
if bin[y][x] {
bm[gy*GridW+gx] = 1
}
}
}
return bm
}
// MatchDigit 与 10 个模板比 Hamming 距离,返回最像的数字。
func MatchDigit(bm Bitmap) int {
best, bestD := 0, 1<<30
for d := 0; d < 10; d++ {
dist := hamming(bm, Templates[d])
if dist < bestD {
bestD = dist
best = d
}
}
return best
}
func hamming(a, b Bitmap) int {
n := 0
for i := range a {
if a[i] != b[i] {
n++
}
}
return n
}