Adapting JPEG2000 bit allocation to preserve features of interest

Nowadays, the demands for higher quality and flexibility in accessing compressed images has led to a gradual replacement of the ubiquitous JPEG standard by the recent JPEG2000 compression standard in various domains. Similar to other compression standards, this new algorithm tends to deteriorate some of the intrinsic geometrical features of the image. For example, the direction of the edges are not well conserved after compression. This papers shows that the optimization metric generally used in JPEG2000, the mean squared error (MSE), is responsible for the non-uniform deterioration of the shapes' angle. To address this problem, a general method of modifying the MSE into any other additive metric is presented and applied to the conservation of underlying shapes' directional information. This modification is handled in a way that does not change the intrinsic implementation and functionalities of any JPEG2000 encoder. This general method of balancing the shapes' deterioration builds on the definition of a directional metric and the computation of edge direction based on a modified Hough Transform, which allows to determine the shapes' slopes with an error of less than the degree, even with low PSNR pictures (up to 15dB). Finally, it is illustrated how our contribution increases the performance of the JPEG2000, when all singularities of an image must be conserved in spite of high compression rates.