OCR Recognition

Overview

The OCR module performs image recognition
The OCR module object prefix is ocr, e.g. ocr.initOcr()
Current OCR engines include ocrLite, TesseractOCR, paddleocr, and paddleOcrOnnxV4

ocr.newOcr Create an OCR Instance

Create an OCR instance

1
function main() {
2
    let o = ocr.newOcr();
3
    // Initialize and recognize here
4
    o.releaseAll()
5
}

ocr.initOcr Initialize

Initialize the OCR module
@param map parameter map
Keys:
type: OCR type — paddleOcrOnnxV4 = PPOCR-V4 model, paddleOcrOnnxV5 = PPOCR-V5 model, paddleOcrOnnxV6 = PPOCR-V6_small model, ocrLite = ocrLite module, tess = tesseractOcr, paddleOcr = Baidu PaddleOCR
If type is ocrLite
Example: {"type":"ocrLite","cpuType:"","baseDir":""},
baseDir: ocrLite library path; usually in the same directory as EC; folder name: OcrLiteNcnn
cpuType: host CPU type — win-lib-cpu-x64, win-lib-cpu-x86, Linux-Lib-CPU, Darwin-Lib-CPU
single: setting to 0 means each device gets its own OCR instance; default is 1; try changing this on different machines
matMode: 1 = convert the image to OpenCV mat in memory and pass it to OCR; 0 = use a file-based approach without conversion
If baseDir and cpuType are omitted, the program finds them automatically
If type is baiduOnline

Example: {"type":"baiduOnline","ak":"xxx","sk":"xx"}

If type is paddleNcnnOcrV5, configure parameters as follows
numThread: thread count; -1 = all, -2 = half of device CPU, 0 = do not set; adjust recognition speed with this value
modelsDir: model file path; built-in models are used if omitted
padding: white border around the image to improve accuracy; increase when the text box does not cover all text; default 32; affects recognition speed
maxSideLen: if the image’s longest edge exceeds maxSideLen, scale it down proportionally; default 640; affects recognition speed
keysName: training label file name; can be external, e.g. keys.txt; built-in model is used if omitted
detName: detection model filename (.param); place under modelsDir; for det.param write det without the param suffix; built-in model is used if omitted
recName: recognition model filename (.onnx); place under modelsDir; built-in model is used if omitted
If type is paddleOcrOnnxV4 or paddleOcrOnnxV5,
Example: {"type":"paddleOcrOnnxV5","single":0,"cpuType:"","baseDir":"","matModel":1},
baseDir: ocrLite library path; usually in the same directory as EC; folder name: PaddleOcrOnnx; optional
numThread: number of CPU cores; default is 1; optional
single: setting to 0 means each device gets its own OCR instance; default is 0; try changing this on different machines; optional
matMode: 1 = convert the image to OpenCV mat in memory to save memory and pass it to OCR; 0 = use a file-based approach without conversion; optional
cpuType: host CPU type — win-lib-cpu-x64, win-lib-cpu-x86, Linux-Lib-CPU, Darwin-Lib-CPU; optional
keysName: training label file path; place under the models folder in baseDir; built-in model is used if omitted
detName: detection model filename (.onnx); place under the models folder in baseDir; built-in model is used if omitted
recName: recognition model filename (.onnx); place under the models folder in baseDir; built-in model is used if omitted
clsName: classification model filename (.onnx); place under the models folder in baseDir; built-in model is used if omitted
padding: white border around the image to improve accuracy; increase when the text box does not cover all text; default 10.
boxThresh: threshold for separating text from background; larger values shrink the text region; range [0, 1]; default 0.3.
boxScoreThresh: threshold for keeping detected text boxes; larger values reduce recall; range [0, 1]; default 0.6.
unClipRatio: controls text box size; larger values make boxes bigger; range [1.6, 2.0]; default 2.0.
doAngleFlag: enable (1)/disable (0) text orientation detection; only needed for upside-down images (rotated 90–270°); default 1
mostAngleFlag: enable (1)/disable (0) angle voting (recognize the whole image using the most likely text orientation); has no effect when doAngleFlag is disabled; default 1
maxSideLen: if the image’s longest edge exceeds maxSideLen, scale it down proportionally; default 0
If type is paddleOcrOnnxV6
Built-in PP-OCRv6_small (det/rec ONNX), coexists with paddleOcrOnnxV5; cls reuses v5
Same parameter fields as paddleOcrOnnxV5; set type to paddleOcrOnnxV6, e.g.: {"type":"paddleOcrOnnxV6","numThread":2,"single":0,"matMode":1}
Default models (optional): PP-OCRv6_small_det.onnx / PP-OCRv6_small_rec.onnx / ppocrv6_small_labels.txt
Faster: for everyday upright UI, use maxSideLen:640, doAngleFlag:0, mostAngleFlag:0
If type is tess
Example: set the Tesseract install path and tessdata path,
Example: {"type":"tess","baseDir:"d:\\tesseract-ocr","path":"d:\\tesseract-ocr\\tessdata","language":"chi_sim","ocrEngineMode":3}
- baseDir: Tesseract install path; download from https://github.com/tesseract-ocr/tesseract/releases or the official site; jTessBoxEditor.zip includes training tools and Tesseract DLLs
- path: Tesseract tessdata folder
- language: language data file; e.g. chi_sim.traineddata is Simplified Chinese, use chi_sim; combine with +, e.g. chi_sim+eng+num
- ocrEngineMode: recognition engine type — 0 OEM_TESSERACT_ONLY, 1 OEM_LSTM_ONLY, 2 OEM_TESSERACT_LSTM_COMBINED, 3 OEM_DEFAULT
- rilLevel: PageIteratorLevel — -1 adaptive, 0 RIL_BLOCK, 1 RIL_PARA, 2 RIL_TEXTLINE, 3 RIL_WORD, 4 RIL_SYMBOL
If type is paddleOcr
Example

1
{
2
  "type": "paddleOcr",
3
  "ocrType":"ONNX_PPOCR_V3",
4
  "padding": 50,
5
  "maxSideLen": 0,
6
  "boxScoreThresh": 0.5,
7
  "boxThresh": 0.3,
8
  "unClipRatio": 1.6,
9
  "doAngleFlag": 0,
10
  "mostAngleFlag": 0
11
}

1
  ocrType : ONNX_PPOCR_V3, ONNX_PPOCR_V4, NCNN_PPOCR_V3
2
  serverUrl: Paddle OCR server address; deploy on another PC and connect from the control center; default is http://127.0.0.1:9022; change the IP when deployed elsewhere; keep the port
3
  padding: white border around the image to improve accuracy; increase when the text box does not cover all text; default 50.<br/>
4
  maxSideLen: scale by the image's long edge; enlarging increases recognition time but improves accuracy; shrinking reduces time but lowers accuracy; 0 means no scaling.<br/>
5
  boxScoreThresh: text box confidence threshold; decrease when the text box does not cover all text <br/>
6
  boxThresh: same purpose; tune experimentally.<br/>
7
  unClipRatio: multiplier for individual text box size; larger values make each box bigger.<br/>
8
  doAngleFlag: enable (1)/disable (0) text orientation detection; only needed for upside-down images (rotated 90–270°); default off.<br/>
9
  mostAngleFlag: enable (1)/disable (0) angle voting; has no effect when doAngleFlag is disabled; default off.<br/>
10
  front: console (1)/tray mode (0); default off.<br/>
11
  daemon: keep OCR service process alive (1)/no (0); default off.<br/>
12
  limit: OCR requests per second; default 1000; lower to reduce CPU usage<br/>
13
  checkImage: verify input is an image (1 yes, 0 no); default off.<br/>

@return {boolean} boolean — success or failure

Tesseract OCR Example

1
function main() {
2
    let tess = {"type": "tess", "path": "d:/tesseract-ocr/tessdata", "baseDir": "d:\\tesseract-ocr"}
3
    // ocrlite
4
    let ocrLite = {
5
        "type": "ocrLite",
6
        "baseDir": "c:/ec/OcrLiteNcnn",
7
        "single": 0,
8
        "cpuType": "win-lib-cpu-x64"
9
    }
10

11

12
    // To use paddleOcr, adjust the parameters yourself
13
    let paddleOcr = {
14
        "type": "paddleOcr",
15
        "ocrType": "ONNX_PPOCR_V3"
16
    }
17

18
    let tocr = ocr.newOcr()
19
    let inited = tocr.initOcr(ocrLite)
20
    logd("初始化结果 -" + inited);
21
    if (!inited) {
22
        loge("error : " + tocr.getErrorMsg());
23
        return;
24
    }
25

26
    for (var ix = 0; ix < 20; ix++) {
27

28
        // Read a bitmap
29
        let bitmap = image.readBitmap("D:/Screenshot_20210127_152932_com.huawei.android.lau.jpg");
30
        if (!bitmap) {
31
            loge("读取图片失败");
32
            continue;
33
        }
34
        console.time("1")
35
        logd("start---ocr");
36
        // Recognize the image
37
        let result = tocr.ocrBitmap(bitmap, 30 * 1000, {"matMode": 1});
38
        logd(result)
39
        if (result) {
40
            logd("ocr结果-》 " + JSON.stringify(result));
41
            for (var i = 0; i < result.length; i++) {
42
                var value = result[i];
43
                logd("文字 : " + value.label + " x: " + value.x + " y: " + value.y + " width: " + value.width + " height: " + value.height);
44
            }
45
        } else {
46
            logw("未识别到结果");
47
        }
48

49
        logd("耗时: " + console.timeEnd(1) + " ms")
50
        image.recycle(bitmap)
51
        sleep(1000);
52
        logd("ix = " + ix)
53
    }
54
    // Release all resources
55
    // paddleOcr closes the OCR process; skip this if you do not need to shut down
56
    tocr.releaseAll();
57
}
58

59
main();

PaddleOcrOnnx OCR Example

1
function main() {
2

3
    // To use paddleOcrOnnxV4, adjust the parameters yourself
4
    // You can also change to paddleOcrOnnxV5 to use PPOCR-V5
5
    let paddleOcrOn = {
6
        "type": "paddleOcrOnnxV4",
7
        "numThread":1,
8
        "single": 0,
9
        "matMode": 1
10
    }
11

12
    let tocr = ocr.newOcr()
13
    let inited = tocr.initOcr(paddleOcrOn)
14
    logd("初始化结果 -" + inited);
15
    if (!inited) {
16
        loge("error : " + tocr.getErrorMsg());
17
        return;
18
    }
19

20
    for (var ix = 0; ix < 20; ix++) {
21

22
        // Capture screenshot
23
        let bitmap = image.captureFullScreen();
24
        if (!bitmap) {
25
            loge("读取图片失败");
26
            continue;
27
        }
28
        console.time("1")
29
        logd("start---ocr "+bitmap);
30
        // Recognize the image
31
        let result = tocr.ocrImage(bitmap, 30 * 1000, {"matMode": 1});
32
        logd(result)
33
        if (result) {
34
            logd("ocr结果-》 " + JSON.stringify(result));
35
            for (var i = 0; i < result.length; i++) {
36
                var value = result[i];
37
                logd("文字 : " + value.label +" confidence:"+value.confidence + " range: " + value.x + "," + value.y + "," + (value.width+value.x) + "," + (value.height+value.y));
38
            }
39
        } else {
40
            logw("未识别到结果");
41
        }
42

43
        logd("耗时: " + console.timeEnd(1) + " ms")
44
        image.recycle(bitmap)
45
        sleep(2000);
46
        logd("ix = " + ix)
47
    }
48
    // Release all resources
49
    // Skip this if you do not need to shut down
50
    tocr.releaseAll();
51
}
52

53
main();

paddleOcrOnnxV6 Example (PP-OCRv6_small)

Use type paddleOcrOnnxV6; built-in PP-OCRv6_small; coexists with paddleOcrOnnxV5
Same parameter fields as paddleOcrOnnxV5; uses control center built-in models when model paths are omitted
Faster: for everyday upright UI, use maxSideLen:640, doAngleFlag:0, mostAngleFlag:0

1
function main() {
2
    let paddleOcrOnnxV6Fast = {
3
        "type": "paddleOcrOnnxV6",
4
        "numThread": 2,
5
        "single": 0,
6
        "matMode": 1,
7
        "padding": 32,
8
        "maxSideLen": 640,
9
        "doAngleFlag": 0,
10
        "mostAngleFlag": 0
11
    }
12

13
    ocr.releaseAll()
14
    let tocr = ocr.newOcr()
15
    let inited = tocr.initOcr(paddleOcrOnnxV6Fast)
16
    logd("初始化结果 -" + inited)
17
    if (!inited) {
18
        loge("error : " + tocr.getErrorMsg())
19
        return
20
    }
21

22
    for (var ix = 0; ix < 20; ix++) {
23
        let bitmap = image.captureFullScreen()
24
        if (!bitmap) {
25
            loge("读取图片失败")
26
            continue
27
        }
28
        console.time("1")
29
        let result = tocr.ocrImage(bitmap, 30 * 1000, {
30
            "padding": 32,
31
            "numThread": 2,
32
            "maxSideLen": 640,
33
            "doAngleFlag": 0,
34
            "mostAngleFlag": 0
35
        })
36
        if (result) {
37
            logd("ocr结果-》 " + JSON.stringify(result))
38
            for (var i = 0; i < result.length; i++) {
39
                var value = result[i]
40
                logd("文字 : " + value.label
41
                    + " confidence:" + value.confidence
42
                    + " range: " + value.x + "," + value.y + ","
43
                    + (value.x + value.width) + "," + (value.y + value.height))
44
            }
45
        } else {
46
            logw("未识别到结果")
47
        }
48
        logd("耗时: " + console.timeEnd(1) + " ms")
49
        image.recycle(bitmap)
50
        sleep(2000)
51
        logd("ix = " + ix)
52
    }
53
    tocr.releaseAll()
54
}
55

56
main()

PaddleOcrNcnnV5 OCR Example [after 2.12.0]

1
function main() {
2

3
    let paddleOcrOn = {
4
        "type": "paddleOcrNcnnV5",
5
        "numThread":1,
6
        "single": 0,
7
        "matMode": 1
8
    }
9

10
    // Release first to avoid leftover resources from a previous session
11
    ocr.releaseAll()
12
    let tocr = ocr.newOcr()
13
    let inited = tocr.initOcr(paddleOcrOn)
14
    logd("初始化结果 -" + inited);
15
    if (!inited) {
16
        loge("error : " + tocr.getErrorMsg());
17
        return;
18
    }
19

20
    for (var ix = 0; ix < 20; ix++) {
21

22
        // Capture screenshot
23
        let bitmap = image.captureFullScreen();
24
        if (!bitmap) {
25
            loge("读取图片失败");
26
            continue;
27
        }
28
        console.time("1")
29
        logd("start---ocr "+bitmap);
30
        // Recognize the image
31
        let result = tocr.ocrImage(bitmap, 30 * 1000, {"padding": 32,"numThread":1});
32
        logd(result)
33
        if (result) {
34
            logd("ocr结果-》 " + JSON.stringify(result));
35
            for (var i = 0; i < result.length; i++) {
36
                var value = result[i];
37
                logd("文字 : " + value.label +" confidence:"+value.confidence + " range: " + value.x + "," + value.y + "," + (value.width+value.x) + "," + (value.height+value.y));
38
            }
39
        } else {
40
            logw("未识别到结果");
41
        }
42

43
        logd("耗时: " + console.timeEnd(1) + " ms")
44
        image.recycle(bitmap)
45
        sleep(2000);
46
        logd("ix = " + ix)
47
    }
48
    // Release all resources
49
    // Skip this if you do not need to shut down
50
    tocr.releaseAll();
51
}
52

53
main();

ocr.ocrBitmap Recognize Text

Perform OCR on a BufferedImage; returns JSON data similar to:

1
[
2
  {
3
    "label": "奇趣装扮三阶盘化",
4
    "confidence": 0.48334712,
5
    "x": 11,
6
    "y": 25,
7
    "width": 100,
8
    "height": 100
9
  }
10
]

label: recognized text
confidence: recognition confidence
x: X start coordinate
Y: Y start coordinate
width: width
height: height
@param bitmap image
@param timeout timeout in milliseconds
@param extra extra parameters as a map, e.g. {"token":"xxx"}
@return {json} JSON object

ocr.ocrImage Recognize Text

Perform OCR on an AutoImage; returns JSON data similar to:

1
[
2
 {
3
 "label": "奇趣装扮三阶盘化",
4
 "confidence": 0.48334712,
5
 "x": 11,
6
 "y": 25,
7
 "width": 100,
8
 "height": 100
9
 }
10
]

label: recognized text
confidence: recognition confidence
x: X start coordinate
Y: Y start coordinate
width: width
height: height
@param bitmap image
@param timeout timeout in milliseconds
@param extra extra parameters as a map, e.g. {"token":"xxx"}
@return {json} JSON object

1
See common examples in
2
OCR initialization

ocr.getErrorMsg Get Error Message

Get the OCR error message
@return {string} null means no error

1
See common examples in
2
OCR initialization

ocr.releaseAll Release OCR Resources

Release OCR resources
@return {bool} success or failure

1
See common examples in
2
OCR initialization