I am currently doing foreground segmentation in image sequences. A DCT 8 by 8 feature is extracted for each pixel.

Now i want to explore dynamic block size for different scenes. Is there any rules or criterion to decide optimal block size?

Thank you!


I assume the block size is related to image quality no matter you are doing the segmentation or compression. Generally speaking, smaller blocks are better as far as correlation (the pixels in the block are more correlated), but if they're too small, you won't be able to eliminate many coefficients from each block without causing severe distortion, especially when there are high variance and more details in the block. Some compression algorithms use different block sizes within the same image so that high-detail regions are divided into smaller blocks, while smooth regions are divided into larger blocks (Chong et al's Compression of medical images through adaptive block-size DCT coding, and The Influence of DCT Block Size on Coding Efficiency: An Adaptive Variable Block Size Transform Coding System with Lagrangian Cost Function Decision Criterion).

Besides, Karunasekera et al's A distortion measure for blocking artifacts in images based on human visual sensitivity provides you a reference to estimate the distortion induced by blocking artifacts. Pramod el al's Unsupervised Segmentation of Medical Images using DCT Coefficients applied DCT coefficients to the image segmentation.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.