diff options
author | Nick White <git@njw.name> | 2019-01-23 20:47:33 +0000 |
---|---|---|
committer | Nick White <git@njw.name> | 2019-01-23 20:47:33 +0000 |
commit | d256f967a26ceeb7c3987a1fc447b126a35054f9 (patch) | |
tree | f80c1e3e2c3757c59ad51dec98de5b6e82a426fa /parse | |
parent | c41aa16d8a3d35ce4185184ee50536bf2089a120 (diff) |
Separate out hocr parts from line parts
Diffstat (limited to 'parse')
-rw-r--r-- | parse/hocr/hocr.go | 147 | ||||
-rw-r--r-- | parse/line.go | 51 |
2 files changed, 198 insertions, 0 deletions
diff --git a/parse/hocr/hocr.go b/parse/hocr/hocr.go new file mode 100644 index 0000000..a281a7a --- /dev/null +++ b/parse/hocr/hocr.go @@ -0,0 +1,147 @@ +package hocr + +// TODO: consider making GetLineDetails() a function of Hocr, so could do a +// similar thing with prob format files too. +// TODO: Parse line name to zero pad line numbers, so they come out in the correct order + +import ( + "encoding/xml" + "image" + "regexp" + "strconv" + "strings" + + "git.rescribe.xyz/testingtools/parse" +) + +type Hocr struct { + Lines []OcrLine `xml:"body>div>div>p>span"` +} + +type OcrLine struct { + Class string `xml:"class,attr"` + Id string `xml:"id,attr"` + Title string `xml:"title,attr"` + Words []OcrWord `xml:"span"` + Text string `xml:",chardata"` +} + +type OcrWord struct { + Class string `xml:"class,attr"` + Id string `xml:"id,attr"` + Title string `xml:"title,attr"` + Chars []OcrChar `xml:"span"` + Text string `xml:",chardata"` +} + +type OcrChar struct { + Class string `xml:"class,attr"` + Id string `xml:"id,attr"` + Title string `xml:"title,attr"` + Chars []OcrChar `xml:"span"` + Text string `xml:",chardata"` +} + +// Returns the confidence for a word based on its x_wconf value +func wordConf(s string) (float64, error) { + re, err := regexp.Compile(`x_wconf ([0-9.]+)`) + if err != nil { + return 0.0, err + } + conf := re.FindStringSubmatch(s) + return strconv.ParseFloat(conf[1], 64) +} + +func boxCoords(s string) ([4]int, error) { + var coords [4]int + re, err := regexp.Compile(`bbox ([0-9]+) ([0-9]+) ([0-9]+) ([0-9]+)`) + if err != nil { + return coords, err + } + coordstr := re.FindStringSubmatch(s) + for i := range coords { + c, err := strconv.Atoi(coordstr[i+1]) + if err != nil { + return coords, err + } + coords[i] = c + } + return coords, nil +} + +func noText(s string) bool { + t := strings.Trim(s, " \n") + return len(t) == 0 +} + +func Parse(b []byte) (Hocr, error) { + var hocr Hocr + + err := xml.Unmarshal(b, &hocr) + if err != nil { + return hocr, err + } + + return hocr, nil +} + +func GetLineDetails(h Hocr, i image.Image, name string) (parse.LineDetails, error) { + lines := make(parse.LineDetails, 0) + + for _, l := range h.Lines { + totalconf := float64(0) + num := 0 + for _, w := range l.Words { + c, err := wordConf(w.Title) + if err != nil { + return lines, err + } + num++ + totalconf += c + } + + coords, err := boxCoords(l.Title) + if err != nil { + return lines, err + } + + var line parse.LineDetail + line.Name = l.Id + line.Avgconf = (totalconf/float64(num)) / 100 + linetext := "" + + linetext = l.Text + if(noText(linetext)) { + linetext = "" + for _, w := range l.Words { + if(w.Class != "ocrx_word") { + continue + } + linetext += w.Text + " " + } + } + if(noText(linetext)) { + linetext = "" + for _, w := range l.Words { + if(w.Class != "ocrx_word") { + continue + } + for _, c := range w.Chars { + if(c.Class != "ocrx_cinfo") { + continue + } + linetext += c.Text + } + linetext += " " + } + } + line.Text = strings.TrimRight(linetext, " ") + line.Text += "\n" + line.Hocrname = name + var imgd parse.ImgDirect + imgd.Img = i.(*image.Gray).SubImage(image.Rect(coords[0], coords[1], coords[2], coords[3])) + line.Img = imgd + lines = append(lines, line) + } + return lines, nil +} diff --git a/parse/line.go b/parse/line.go new file mode 100644 index 0000000..3ddde76 --- /dev/null +++ b/parse/line.go @@ -0,0 +1,51 @@ +package parse + +// TODO: integrate in line-conf-buckets linedetail +// TODO: add BucketUp() function here that does what both line-conf-buckets-tess.go +// and line-conf-buckets.go do +// TODO: consider naming this package line, and separating it from hocr and prob + +import ( + "image" + "image/png" + "io" +) + +type LineDetail struct { + Name string + Avgconf float64 + Img CopyableLine + Text string + Hocrname string +} + +type CopyableLine interface { + CopyLineTo(io.Writer) (error) +} + +// This is an implementation of the CopyableLine interface that +// stores the image directly as an image.Image +type ImgDirect struct { + Img image.Image +} + +func (i ImgDirect) CopyLineTo(w io.Writer) (error) { + err := png.Encode(w, i.Img) + if err != nil { + return err + } + return nil +} + +type LineDetails []LineDetail + +// Used by sort.Sort. +func (l LineDetails) Len() int { return len(l) } + +// Used by sort.Sort. +func (l LineDetails) Less(i, j int) bool { + return l[i].Avgconf < l[j].Avgconf +} + +// Used by sort.Sort. +func (l LineDetails) Swap(i, j int) { l[i], l[j] = l[j], l[i] } |