fin-ocr-cli generates bad images

[aaronreed708]

Bug Report

Steps to Reproduce:

1. clone and build fin-ocr-cli
2. run ./scripts/ocr check generate 1
3. run ./scripts/ocr check scan ~/.fin-ocr/checks/check-1.png

Expected Result:

User should see a JSON data structures containing the generated check information. Something like:

{
    "id": "/Users/aaron/.fin-ocr/checks/check-1.png",
    "translators": {
        "tesseract": {
            "result": {
                "micrLine": "[aoe 2 ete eee ie. - Lee Se\n",
                "routingNumber": "",
                "accountNumber": "",
                "checkNumber": ""
            }
        },
        "opencv": {
            "result": {
                "micrLine": "T751116251T259233723U4437",
                "routingNumber": "751116251",
                "accountNumber": "259233723",
                "checkNumber": "4437"
            }
        }
    },
    "overlap": true
}

Actual Result:

Sending scan request for ID: /Users/aaron/.fin-ocr/checks/check-1.png
Error processing file /Users/aaron/.fin-ocr/checks/check-1.png: Error: Failed to decode image/png: unrecognised content at end of stream
    at Util.bufferToMat (file:///Users/aaron/crap/fin-ocr-sdk/build/src/util.js:38:19)
    at Image.fromBuffer (file:///Users/aaron/crap/fin-ocr-sdk/build/src/image.js:52:26)
    at Check.getMicrLine (file:///Users/aaron/crap/fin-ocr-sdk/build/src/check.js:158:25)
    at Check.scan (file:///Users/aaron/crap/fin-ocr-sdk/build/src/check.js:41:37)
    at CheckMgr.getScanResponse (file:///Users/aaron/crap/fin-ocr-cli/build/check.js:271:42)
    at CheckMgr.scan (file:///Users/aaron/crap/fin-ocr-cli/build/check.js:203:31)
    at checkScan (file:///Users/aaron/crap/fin-ocr-cli/build/main.js:134:41)
    at async check (file:///Users/aaron/crap/fin-ocr-cli/build/main.js:51:13)
    at async main (file:///Users/aaron/crap/fin-ocr-cli/build/main.js:32:9)

Environment:

Issue noticed on fin-ocr-cli main yesterday using fin-ocr-sdk@1.0.5

[aaronreed708]

I did some problem determination on this. Failure seems to have been introduced with commit #1be3991

Looking at the generated file, when it is "bad" I see:

Aarons-MacBook-Pro:fin-ocr-cli aaron$ xxd ~/.fin-ocr/checks/check-1.png
00000000: efbf bd50 4e47 0d0a 1a0a 0000 000d 4948  ...PNG........IH
00000010: 4452 0000 0258 0000 00ef bfbd 0806 0000  DR...X..........

When it is "good", I see:

Aarons-MacBook-Pro:fin-ocr-cli aaron$ xxd ~/.fin-ocr/checks/check-1.png
00000000: 8950 4e47 0d0a 1a0a 0000 000d 4948 4452  .PNG........IHDR
00000010: 0000 0258 0000 00fa 0806 0000 007d 2d39  ...X.........}-9

According to https://stackoverflow.com/questions/50910874/unexpected-buffer-result-when-fetching-png-image, the hex value differences could be because the encoding is incorrect. As far as I can tell, generateCheckImage didn't change in the patch in question. But the file write did change:

public async generateCheckImages(count: number): Promise<void> {
        // Create checks directory if it doesn't exist
        if (!fs.existsSync(this.checksDir)) {
            fs.mkdirSync(this.checksDir, { recursive: true });
        }

        for (let i = 1; i <= count; i++) {
            const filePath = path.join(this.checksDir, `check-${i}.png`);
            const imageData = await this.generateCheckImage(i);
            fs.writeFileSync(filePath, imageData.toString());                           <----------- change here added toString()
            ctx.info(`Generated check image: ${filePath}`);
        }
        if (fs.existsSync('temp.png')) {
            fs.unlinkSync('temp.png');
            ctx.info(`Deleted temporary file: ${'temp.png'}`);
        }
    }

I think that the fix would be to not use toString() here (which has a default encoding of utf-8). Making that change fixes the problem. However, I have no idea if this will re-expose the problem that the patch hoped to originally fix.

[aaronreed708]

Based on my notes at the time and the commit message, I'm guessing that it is related to: https://github.com/finos/fin-ocr-cli?tab=readme-ov-file#how-to-use-the-cli-as-a-client-for-the-rest-service