Extending attribute filters to color processing and multi-media applications

PhD ceremony: Ms. F.B. Tushabe, 14.45 uur, Academiegebouw, Broerstraat 5, Groningen

Thesis: Extending attribute filters to color processing and multi-media applications

Promotor(s): prof. N. Petkov

Faculty: Mathematics and Natural Sciences

In this work, connected attribute filters are implemented using the Max-tree approach and extended to aid color image processing using two new approaches. The first one builds the tree based on luminance, saturation, chromaticity or their combinations. The image is then filtered and the color image reconstructed by assigning the remaining structures their original colors. When these methods are applied to a traffic signs recognition problem, the results show an automatic sign recognition of 85%.

The second approach reconstructs the color image using three new restitution rules which have all resulted in an improvement in color fidelity. Comparison with an earlier method show that the proposed methods result in quality improvement by as much as 15%.

This work also proposes filters that improve compression results in terms of file size and image quality. They remove psycho-visually redundant data from an image by mimicking how the human visual system works and have been found to modify as much as 35 – 40% of the image contents without causing visual losslessness.

A filter that improves content-base image retrieval for general purpose vacation pictures is also proposed. It has a high object discriminating power and is scale, rotation and translation invariant. Finally, the effect of applying five attribute filters on a watermarked image has been investigated. When subjected to seven watermarking algorithms, 92% of the filtered images detected the embedded watermarks.This means that attribute filtering is a largely safe and robust procedure as a pre-processing method for any given application.