summaryrefslogtreecommitdiff
path: root/line-conf-buckets-tess/line-conf-buckets-tess.go
blob: 29df9f47ea2bb293569fc3507fe38f8a2304c290 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
package main

// TODO: combine this with line-conf-buckets, separating the parsing
//       out to a separate library, probably
// see https://github.com/OCR-D/ocrd-train/issues/7 and https://github.com/OCR-D/ocrd-train/
// for tips on creating lines of tif/txt. best thing is to use hocr-extract-images to extract
// images for each line, based on tesseract's hocr output. can then copy the ground truth
// for that
// initial plan for this is to identify the lines which are best, and extract the text, then
// later can extract the images from them
//
// ok, am parsing the hocr now, workflow should be:
// - run hocr-extract-images (outside of this) and have a directory of images named line-000.png
// - run this with hocr and hocr-images dir
//   this then saves the text for the line alongside copying the image from the dir into a fresh dir, according to the line confidence
//
// actually, *should* be able to extract the images quite straightforwardly straight from go, which would be cool. so try to build that.
// should be super easy, with SubImage, see end of https://blog.golang.org/go-image-package

import (
	"encoding/xml"
	"flag"
	"fmt"
	"image"
	"image/png"
	"io"
	"io/ioutil"
	"log"
	"os"
	"path/filepath"
	"regexp"
	"sort"
	"strconv"
	"strings"
)

type LineDetail struct {
	name string
	avgconf float64
	img image.Image
	text string
	hocrname string
}

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] }

func copyline(filebase string, dirname string, basename string, avgconf string, outdir string, todir string) (err error) {
	outname := filepath.Join(outdir, todir, filepath.Base(dirname) + "_" + basename + "_" + avgconf)
	//log.Fatalf("I'd use '%s' as outname, and '%s' as filebase\n", outname, filebase)

	for _, extn := range []string{".bin.png", ".txt"} {
		infile, err := os.Open(filebase + extn)
		if err != nil {
			fmt.Fprintf(os.Stderr, "Failed to open %s\n", filebase + extn)
			return err
		}
		defer infile.Close()

		err = os.MkdirAll(filepath.Join(outdir, todir), 0700)
		if err != nil {
			return err
		}
	
		outfile, err := os.Create(outname + extn)
		if err != nil {
			fmt.Fprintf(os.Stderr, "Failed to create %s\n", outname + extn)
			return err
		}
		defer outfile.Close()
	
		_, err = io.Copy(outfile, infile)
		if err != nil {
			return err
		}
	}

	return err
}

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"`
	// TODO: also capture OcrChar where it exists, to grab text from it
	// TODO: grab text from these elements, to save for the line
}

// 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 main() {
	flag.Usage = func() {
		fmt.Fprintf(os.Stderr, "Usage: line-conf-buckets hocr1 [hocr2] [...]\n")
		fmt.Fprintf(os.Stderr, "Copies image-text line pairs into different directories according\n")
		fmt.Fprintf(os.Stderr, "to the average character probability for the line.\n")
		flag.PrintDefaults()
	}
	flag.Parse()
	if flag.NArg() < 1 {
		flag.Usage()
		os.Exit(1)
	}

	lines := make(LineDetails, 0)

	var hocr Hocr

	for _, f := range flag.Args() {
		file, err := ioutil.ReadFile(f)
		if err != nil {
			log.Fatal(err)
		}

		err = xml.Unmarshal(file, &hocr)
		if err != nil {
			log.Fatal(err)
		}

		pngfn := strings.Replace(f, ".hocr", ".png", 1)
		pngf, err := os.Open(pngfn)
		if err != nil {
			log.Fatal(err)
		}
		defer pngf.Close()
		img, err := png.Decode(pngf)
		if err != nil {
			log.Fatal(err)
		}

		for _, l := range hocr.Lines {
			totalconf := float64(0)
			num := 0
			for _, w := range l.Words {
				c, err := wordConf(w.Title)
				if err != nil {
					log.Fatal(err)
				}
				num++
				totalconf += c
			}

			coords, err := boxCoords(l.Title)
			if err != nil {
				log.Fatal(err)
			}

			var line LineDetail
			line.name = l.Id
			line.avgconf = totalconf/float64(num)
			line.text = l.Text // TODO: get text from OcrWord and OcrChar (if available)
			line.hocrname = strings.Replace(filepath.Base(f), ".hocr", "", 1)
			line.img = img.(*image.Gray).SubImage(image.Rect(coords[0], coords[1], coords[2], coords[3]))
			lines = append(lines, line)
		}
	}

	sort.Sort(lines)

	worstnum := 0
	mediumnum := 0
	bestnum := 0

	outdir := "buckets" // TODO: set this from cmdline
	todir := ""

	for _, l := range lines {
		switch {
		case l.avgconf < 0.95: 
			todir = "bad"
			worstnum++
		case l.avgconf < 0.98:
			todir = "95to98"
			mediumnum++
		default:
			todir = "98plus"
			bestnum++
		}

		avgstr := strconv.FormatFloat(l.avgconf, 'G', -1, 64)
		avgstr = strings.Replace(avgstr, ".", "", 1)
		fmt.Printf("Line: %s, avg: %f, avgstr: %s\n", l.name, l.avgconf, avgstr)
		outname := filepath.Join(outdir, todir, l.hocrname + "_" + l.name + "_" + avgstr + ".png")

		err := os.MkdirAll(filepath.Join(outdir, todir), 0700)
		if err != nil {
			log.Fatal(err)
		}
	
		outfile, err := os.Create(outname)
		if err != nil {
			fmt.Fprintf(os.Stderr, "Failed to create %s\n", outname)
			log.Fatal(err)
		}
		defer outfile.Close()

		err = png.Encode(outfile, l.img)
		if err != nil {
			log.Fatal(err)
		}
		// TODO: do same with saving line

		// TODO: copy the line.img and line.text into the appropriate place, using hocrname/name.ext
		// TODO: test whether the line.img works properly with multiple hocrs, as it could be that as it's a pointer, it always points to the latest image (don't think so, but not sure)
	}

	total := worstnum + mediumnum + bestnum

	fmt.Printf("Copied lines to %s\n", outdir)
	fmt.Printf("---------------------------------\n")
	fmt.Printf("Lines 98%%+ quality:     %d%%\n", 100 * bestnum / total)
	fmt.Printf("Lines 95-98%% quality:   %d%%\n", 100 * mediumnum / total)
	fmt.Printf("Lines <95%% quality:     %d%%\n", 100 * worstnum / total)
}