Farsi (or other non latin characters) not in markdown from image

[chregu]

Question

When converting an image with eg. farsi text in it, it shows up in the Docling Document, but not in the generated markdown (since its parent is a picture, I guess). The latin characters are in the markdown.

Is that to be supposed like that?

Tried with EasyOCR and TesseractCli. And also supplying the correct lang parameters (hopefully). Same also happens, when I supply an image with just Farsi and no latin characters.

I can of course write my own DoclingDocument parser for that, just wondering, why it doesn't show up in the default markdown converter and why it's a picture, while latin characters are not.

This is the image

And this is the produced document.

 {
    "schema_name": "DoclingDocument",
    "version": "1.0.0",
    "name": "docling_4fc1bc1e-5b05-491e-a7a7-4dece794501e_f738b51a5a2c415281ab32b2b42397db-1762493015",
    "origin": {
        "mimetype": "application/pdf",
        "binary_hash": 12687658831323329000,
        "filename": "docling_4fc1bc1e-5b05-491e-a7a7-4dece794501e_f738b51a5a2c415281ab32b2b42397db-1762493015.jpg"
    },
    "furniture": {
        "self_ref": "#/furniture",
        "children": [],
        "name": "_root_",
        "label": "unspecified"
    },
    "body": {
        "self_ref": "#/body",
        "children": [
            {
                "$ref": "#/pictures/0"
            },
            {
                "$ref": "#/texts/1"
            },
            {
                "$ref": "#/texts/2"
            },
            {
                "$ref": "#/texts/3"
            }
        ],
        "name": "_root_",
        "label": "unspecified"
    },
    "groups": [],
    "texts": [
        {
            "self_ref": "#/texts/0",
            "parent": {
                "$ref": "#/pictures/0"
            },
            "children": [],
            "label": "section_header",
            "prov": [
                {
                    "page_no": 1,
                    "bbox": {
                        "l": 15.954619957683569,
                        "t": 834,
                        "r": 1073.544136848707,
                        "b": 584.6887967043365,
                        "coord_origin": "BOTTOMLEFT"
                    },
                    "charspan": [
                        0,
                        39
                    ]
                }
            ],
            "orig": "تو دیگرم شدی شدی شدم من من من تو ازي من",
            "text": "تو دیگرم شدی شدی شدم من من من تو ازي من",
            "level": 1
        },
        {
            "self_ref": "#/texts/1",
            "parent": {
                "$ref": "#/body"
            },
            "children": [],
            "label": "text",
            "prov": [
                {
                    "page_no": 1,
                    "bbox": {
                        "l": 63.33333206176758,
                        "t": 531.3333129882812,
                        "r": 1023.6666870117188,
                        "b": 447,
                        "coord_origin": "BOTTOMLEFT"
                    },
                    "charspan": [
                        0,
                        106
                    ]
                }
            ],
            "orig": "Mun tu shudam tu mun shudimun tun shudam tu jaan shudi Taakas na guyad baad azeen, ٨٧٨ deegaram tu deegari",
            "text": "Mun tu shudam tu mun shudimun tun shudam tu jaan shudi Taakas na guyad baad azeen, ٨٧٨ deegaram tu deegari"
        },
        {
            "self_ref": "#/texts/2",
            "parent": {
                "$ref": "#/body"
            },
            "children": [],
            "label": "text",
            "prov": [
                {
                    "page_no": 1,
                    "bbox": {
                        "l": 147,
                        "t": 411,
                        "r": 928,
                        "b": 250,
                        "coord_origin": "BOTTOMLEFT"
                    },
                    "charspan": [
                        0,
                        131
                    ]
                }
            ],
            "orig": "I have become you, and you I am the you soul; So that no one can say hereafter, That you are are someone, and me someone else body,",
            "text": "I have become you, and you I am the you soul; So that no one can say hereafter, That you are are someone, and me someone else body,"
        },
        {
            "self_ref": "#/texts/3",
            "parent": {
                "$ref": "#/body"
            },
            "children": [],
            "label": "section_header",
            "prov": [
                {
                    "page_no": 1,
                    "bbox": {
                        "l": 412,
                        "t": 205.6666717529297,
                        "r": 664,
                        "b": 162.6666717529297,
                        "coord_origin": "BOTTOMLEFT"
                    },
                    "charspan": [
                        0,
                        12
                    ]
                }
            ],
            "orig": "Ameer Khusro",
            "text": "Ameer Khusro",
            "level": 1
        }
    ],
    "pictures": [
        {
            "self_ref": "#/pictures/0",
            "parent": {
                "$ref": "#/body"
            },
            "children": [
                {
                    "$ref": "#/texts/0"
                }
            ],
            "label": "picture",
            "prov": [
                {
                    "page_no": 1,
                    "bbox": {
                        "l": 23.88103675842285,
                        "t": 831.9660034179688,
                        "r": 1062.5076904296875,
                        "b": 582.7559814453125,
                        "coord_origin": "BOTTOMLEFT"
                    },
                    "charspan": [
                        0,
                        0
                    ]
                }
            ],
            "captions": [],
            "references": [],
            "footnotes": [],
            "annotations": []
        }
    ],
    "tables": [],
    "key_value_items": [],
    "pages": {
        "1": {
            "size": {
                "width": 1080,
                "height": 1080
            },
            "page_no": 1
        }
    }
}

[dolfim-ibm]

Here are some findings:

1. The Farsi text is indeed detected as "content of a picture", hence not exported (by default) in the markdown.
2. The rest of the text instead is not detected as picture.

We have an example (soon to be merged) which shows how to work with the text in pictures: https://github.com/DS4SD/docling/pull/624/files.