Transform coding of images using local orientation adaptivity

Conventional image decomposition techniques are limited in their redundancy reduction properties due to their inability to detect essential structural aspects in images. By using image analysis tools that are capable of detecting and extracting important image structures, however, more efficient coding algorithms can be developed. Also, content-based coding enables compression algorithms to be tuned more specifically to visually important image elements. To demonstrate the efficiency of such a content-based approach, we present an image compression scheme based on a Hermite transform that adapts to local image orientations. Simulations show that orientation adaptivity results in a significant reduction of redundancy. Comparisons with other compression techniques such as JPEG indicate that the proposed scheme performs very well for high compression ratios, not only in terms of peak-signal-tonoise ratio but also in terms of perceptual image quality.