Getting PixelFormat from Jpeg

[christian-be]

Hi!

Does anybody know, if there is a way to determine the PixelFormat of an image (jpg) dynamically using ImageSharp. With JpgBitmapDecoder from WPF, a frame has this information available in the format property. https://docs.microsoft.com/en-us/dotnet/api/system.windows.media.pixelformat?view=netcore-3.1

https://docs.sixlabors.com/api/ImageSharp/SixLabors.ImageSharp.PixelFormats.html

Did not find anything regarding this in the docs. Appreciate any help.

[JimBobSquarePants]

ImageSharp works a little differently...

The PixelFormat is imperative, i.e. determined by the generic type parameter given to Image<TPixel>. The non-generic variant currently defaults to Rgba32 but we may change that in the future as we optimize each decoder.

You can however get the original bit depth from a jpeg via PixelType property on the image. You can use the various Identify methods to get this information without fully decoding the image to should you choose to do so.

Identify and co are super, super fast so you could use them to determine the bit depth and the IImageFormat which would allow you to create optimized paths for each image format and then decode along those paths.

[antonfirsov]

@christian-be can you elaborate on your use case? Why is that information useful? What decisions does your app make based on the pixel format?

[christian-be]

Thanks for the quick replies! We are using ImageSharp to crop images before sending them to an external 3rd-party OCR-engine. This library expects an enum to be passed alongside the image-stream containing this information. (Bgr24 | Bgr32 | Gray8...etc). What I can see from @JimBobSquarePants answer is, that it's easy to detect the pixel-depth of the image, but how to determine if it is RGB, Gray, BW?

[christian-be]

Alright, I just tried James's solution, the OCR-library complains about an unknown image format. So it looks like, they do expect a jpeg. Unfortunately the documentation they provide is not very extensive.

@antonfirsov: You wrote

Current ImageSharp behavior is that the pixel format of your image object does not count at all when you call image.Save(...)

Are there plans for changing this behaviour? I don't have a lot experience with image manipulation, but wouldn't it make sense, to give a client the possibility to explicitly set this? Not sure if there are many usecases, where someone would need to encode an image with a specific pixel format...

[antonfirsov]

You can configure the format of the encoded stream on the custom encoder instance you provide to image.Save(...).

The jpeg standard defines the following formats as internal representation: YCbCr, Ycck, Cmyk, Grayscale, Rgb. We only support the first, the rest is tracked in #808. I don't know how would it help your case, since those formats are only for the jpeg stream, unrelated to the WPF pixel formats, and usually the image gets decoded to different formats by imaging libraries.

I still think the OCR library will most likely ignore this information, since it's redundant and useless.

[antonfirsov]

Ok, I think I have an idea now:
it is likely a hint to the OCR library on what format should it decode into internally. In that case, any 3-component RGB-ish format should be fine (eg WPF's PixelFormat.Rgb24)

[christian-be]

The results from the OCR look good with the current setup, so it looks like you might be right. Maybe they introduced this, to support also more uncommon formats, which would cause problems when decoding assuming there will be 3 channels available.

[antonfirsov]

Btw if image quality matters (eg. to get better recognition rates), it's probably better to pass the image in a lossless format like PNG (new PngEncoder() ), or set set the highest quality on JpegEncoder:

new JpegEncoder() { Quality = 100 }