Error in pixReadFromTiffStream: old style jpeg format is not supported

[milahu]

Current Behavior

tesseract 5.5.1 tries to use the thumbnail images embedded in tiff files
but it fails to open the thumbnail images and shows the warning

Error in pixReadFromTiffStream: old style jpeg format is not supported

test image

helloworld-with-thumb.tiff

wget https://github.com/user-attachments/files/23533251/helloworld-with-thumb.tiff
tesseract -l eng helloworld-with-thumb.tiff -

Page 1
Hello world
Error in pixReadFromTiffStream: old style jpeg format is not supported

Reproduce

gimp -> export as -> test.tiff -> compression: jpeg, save thumbnail
run tesseract test.tiff

Expected Behavior

tesseract should see that these are just thumbnail images
and should silently ignore them

Suggested Fix

No response

tesseract version

$ tesseract -v
tesseract 5.5.1
 leptonica-1.85.0
  libgif 5.2.2 : libjpeg 6b (libjpeg-turbo 3.1.0) : libpng 1.6.47 : libtiff 4.7.0 : zlib 1.3.1 : libwebp 1.5.0 : libopenjp2 2.5.2

Operating System

No response

Other Operating System

No response

uname -a

No response

Compiler

No response

CPU

No response

Virtualization / Containers

No response

Other Information

the warning message comes from leptonica/src/tiffio.c

        /* Old style jpeg is not supported.  We tried supporting 8 bpp.
         * TIFFReadScanline() fails on this format, so we used RGBA
         * reading, which generates a 4 spp image, and pulled out the
         * red component.  However, there were problems with double-frees
         * in cleanup.  For RGB, tiffbpl is exactly half the size that
         * you would expect for the raster data in a scanline, which
         * is 3 * w.  */
    TIFFGetFieldDefaulted(tif, TIFFTAG_COMPRESSION, &tiffcomp);
    if (tiffcomp == COMPRESSION_OJPEG) {
        L_ERROR("old style jpeg format is not supported\n", __func__);
        return NULL;
    }

continue #3008

[Mohataseem89]

Hi I would like to work on this issue.

I have tried to investigated the error and may found that it comes from Leptonica’s tiffio.c, where OJPEG (old-style JPEG) compressed subfiles cause L_ERROR("old style jpeg format is not supported")

In many TIFF files (e.g., GIMP exports), the embedded thumbnail uses OJPEG, not the main image. because thumbnails are marked with TIFFTAG_SUBFILETYPE & FILETYPE_REDUCEDIMAGE, the library can safely skip those without showing an error

Proposed fix:
Check whether the current TIFF subfile is a reduced image
If yes --> return NULL without logging an error
If not --> keep current error behavior

I can prepare a PR with this patch.
let me know if this approach is acceptable.

[amitdo]

In general, we don't want to directly interact with image formats. Instead, we use Leptonica. So this should be solved in the Leptonica code.

@DanBloomberg

[DanBloomberg]

Correct me if I'm wrong. When leptonica tries to read one of these OJPEG compressed files, it returns an error. You would like it to return a WARNING and not an ERROR when it attempts to uncompress such a file if that file is labeled with both TIFFTAG_SUBFILETYPE and FILETYPE_REDUCEDIMAGE.

But the caller is asking to make a Pix. Leptonica can't make the pix, so as you have correctly stated, in all such situations it must return NULL for the data. The caller needs to deal with that.

So please explain how it matters to the caller whether the error message says "Error" or "Warning". The function in both cases returns NULL.

[DanBloomberg]

Handling image formats in applications is a messy, never-ending business. Witness this issue!
As @amitdo correctly stated, tesseract should not get involved with image formats.
And as I've explained, the caller of a leptonica function looks at return values, not messages.
And the return value of an image reading function is either the image or null.

So I do not see how this issue can be resolved in either tesseract or leptonica.
Instead, the OJPEG formatted images either need to be removed or not generated in the first place.
Perhaps gimp has a setting for thumbnail formats?

[amitdo]

Dan,

dentify -verbose        '/home/amit/Downloads/helloworld-with-thumb.tiff'

Image: /home/amit/Downloads/helloworld-with-thumb.tiff
  Format: TIFF (Tagged Image File Format)
  Geometry: 400x100
  Class: DirectClass
  Type: grayscale
  Depth: 8 bits-per-pixel component
  Channel Depths:
    Gray:     8 bits
    Opacity:  1 bits
  Channel Statistics:
    Gray:
      Minimum:                     0.00 (0.0000)
      Maximum:                 65535.00 (1.0000)
      Mean:                    59419.27 (0.9067)
      Standard Deviation:      17466.80 (0.2665)
    Opacity:
      Minimum:                     0.00 (0.0000)
      Maximum:                     0.00 (0.0000)
      Mean:                        0.00 (0.0000)
      Standard Deviation:          0.00 (0.0000)
  Resolution: 300x300 pixels/inch
  Filesize: 21.6Ki
  Interlace: No
  Orientation: TopLeft
  Background Color: white
  Border Color: #DFDFDF00
  Matte Color: #BDBDBD00
  Page geometry: 400x100+0+0
  Compose: Over
  Dispose: Undefined
  Iterations: 0
  Scene: 0 of 2
  Compression: JPEG
  Alpha: Unassociated
  Timestamp: 2025:11:13 20:29:47
  Comment: Created with GIMP
  Label: Background
  Software: GIMP 2.10.38
  Signature: 0be8461212331ec448783ce4da1c7564f7558329c26dc31e08e90c25b48727db
  Profile-color: 672 bytes
  Profile-XMP: 3523 bytes
  Tainted: False
  Elapsed Time: 0m:0.001116s
  Pixels Per Second: 34.2Mi
Image: /home/amit/Downloads/helloworld-with-thumb.tiff
  Format: TIFF (Tagged Image File Format)
  Geometry: 256x64
  Class: DirectClass
  Type: grayscale with transparency
  Depth: 8 bits-per-pixel component
  Channel Depths:
    Gray:     8 bits
    Opacity:  1 bits
  Channel Statistics:
    Gray:
      Minimum:                     0.00 (0.0000)
      Maximum:                 65535.00 (1.0000)
      Mean:                    59768.80 (0.9120)
      Standard Deviation:      16298.58 (0.2487)
    Opacity:
      Minimum:                 65535.00 (1.0000)
      Maximum:                 65535.00 (1.0000)
      Mean:                    65535.00 (1.0000)
      Standard Deviation:          0.00 (0.0000)
  Opacity: (255,255,255,255)	  #FFFFFFFF
  Filesize: 21.6Ki
  Interlace: No
  Orientation: Unknown
  Background Color: white
  Border Color: #DFDFDF00
  Matte Color: #BDBDBD00
  Page geometry: 256x64+0+0
  Compose: Over
  Dispose: Undefined
  Iterations: 0
  Scene: 1 of 2
  Compression: JPEG
  Alpha: Unassociated
  Signature: a6a34d457d3c84cba284205465abc867762baa85ce7cc038b399b2ef3389a9dc
  Tainted: False
  Elapsed Time: 0m:0.000569s
  Pixels Per Second: 27.5Mi

So according to ImageMagick the thumbnail format is JPEG.

[amitdo]

tiffinfo '/home/amit/Downloads/helloworld-with-thumb.tiff'

main image

Compression Scheme: JPEG

thumbnail

Compression Scheme: Old-style JPEG

[amitdo]

Let's ignore the ojpeg issue for a moment.

The general issue is that Tesseract has no way to know that a tiff image has thumbnails. If tesseract knows that a given page is an actually just a thumbnail, it should skip this page and not try to OCR it.