Segmentation-based image compression with enhanced treatment of textured regions

This paper presents a method to preprocess an image so that when segmented it yields a partitioning in which textured regions are approximated with a substantially reduced number of uniform regions (which is desirable for the coding). The segmentation method used to form this representation combines a Gaussian texture model and Gibbs-Markov contour model in order to find regions with boundaries which correspond closely to the objects in the image. Given the image segmentation, an approximation to the original image is generated by filling each region with its mean value. If higher quality reconstruction is desired, the quantized approximation error is also encoded. In order to exploit the reduced sensitivity of the human visual system to the error around edges (visual masking), the error is quantized using three nonlinear quantizers corresponding to the smoothly varying, textured, and remaining areas of the image, respectively