Progressive image transmission by segmentation coding

Abstract

A . coding algorithm is presented for progressive transmission of gray-tone images. The approach is motivated by the desire to retain edge information at low bit rates. A segmentation is first performed using the watershed algorithm to yield an oversegmented mosaic image that is successively decimated. The region contours 8te chain encoded and transmitted in reverse order, while the lowpass information is coded using DPCM.Whenever we want to transmit an image over a low rate channel, it becomes attractive to be able to quickly transmit a crude rendition that can be gradually improved upon. Progressive image transmission (PIT) does just that. Many PIT schemes have been proposed in the past two decades, ranging from bit-plane to vector quantization (VQ), multiresolution pyramids and progressively transmitted transform coefficients [l], with the latest development in that area focusing on the wavelet. With the emergence of large remote databases, PIT has the potential of becoming an important browsing tool. In that case, the amount of information transmitted at the early stages becomes crucial. Since the visual information is condensed in the object contours, this translates as edge preservation.We consider here, segmentation-based coding [2], a technique in which an image is partitioned into regions with homogeneous properties, as a way of achieving this. Compression is attained by encoding the region parameters and the contour of these regions. In this paper, the regions are represented by a piecewise constant approximation. Because of this, a large number of regions are needed to ensure a high quality reconsauction. However, savings can be made on the region parameters during the coding phase. Furthermore, the method can be made adaptive at no cost. as we will show in the next sections. The difficulty in segmentation coding lies mainly in achieving the first segmentation. However, mathematical morphology has a powerful segmentation algorithm: the watershed transformation. Under this transformation, the image is regarded as a topological surface that is gradually flooded [3]. The minima of the image are first sorted. Flooding starts at the basin of the minima at the lowest altitude. When the water level reaches the altitude of the second minimum, both are flooded simultaneously. This procedure is iterated until all the pixels in the image have been reached. At the same time, dams are erected at locations where the water from two or more different basins would merge. These dams are the watersheds (see Figl). water level I Y Figurel. The watersheds To partition an image into uniform regions, we apply the watershed to the gradient of that image. The morphological gradient is preferred for its uniformity. If f(Xi) denotes an image and Mn a window or flat structuring element of size n, the erosion and dilation are given by: Erosion: &n(f)(Xi) = Min(f(Xi+k), kc Mn)The morphological gradient is simply defined as:

Survey of progressive image transmission methods

“…The final transmission involves the texture component. While different segmentation algorithms and coding methods are used in Subramanian et al (1992) and Caron and Rivest (1996), their structures for PIT are essentially similar.…”

“…The design of SIC involves specifying the parts that perform segmentation, contour coding, and region coding (Kwon and Chellappa, 1993). The use of segmented image coders for PIT has been described in Lowe (1993), Subramanian et al (1992), and Caron and Rivest (1996). Refinement of the image is obtained by transmitting the various image components in a sequence.…”

Progressive Image Transmission by Segmentation-Based Coding

“…These second-highly degraded version of the image, and construct the region, at the receiving end, by progressively splitting the generation techniques are better suited for the human visual system, even with the very simple region characteriza-regions already received. We called this method Segmentation-Based Coding (SBC) [4]. tion model we presently use.…”