Constraint adaptive segmentation for color image coding and content-based retrieval

Abstract

-We present a constraint adaptive image segmentation technique designed to achieve the combined purposes of color image coding/compression, indexing and content-based retrieval. An image is segmented into homogeneous squared regions of variable sizes such that it can be encoded very emciently [7]. From the segmented image, we derive an effective and efficient image content description feature termed the region and color co-occurrence matrix (RACOM) as image index for content-based indexing and retrieval. Experimental results are presented to demonstrate that RACOM is a very effective image content description features and has comparable performance to state of the art methods, colour correlogram [8] and MPEG7 colour structure histogram [9], in content-based image retrieval from large image databases.

Embedded colour image coding for content-based retrieval

“…Following the essential idea of Vaisey and Gersho (1992), we have developed a simple implementation of the variable block size segmentation method (Qiu, 2001b). The constraints we used in this work were as follows: (1) the shapes of the regions were restricted to be square; (2) the maximum size of the regions must not exceed N Â N; (3) within each region, the pixels must have similar colours.…”

“…From this set of indices, we can construct a similar image content descriptor to the region and colour co-occurrence matrix (Qiu, 2001b). Let P asp ði; jÞ be the probability of a block of size i (determined by the segmentation algorithm of Section 2) and whose ASP vector is encoded by the vector quantizer to the jth codeword of VQ asp .…”

“…We believe this line of research will be useful in managing large image databases. In particular it can be used to compress images to moderate bite rates and to enable fast processing in the compressed domain for easy content access (Qiu, 2001b). Completely unconnected to image coding and compressed domain content access, the blobworld system (Carson et al, 1999) has demonstrated the advantages of segmenting images into meaningful regions for content-based retrieval.…”

“…In this paper, we present a method which combines the constrained adaptive segmentation-based image coding method (Qiu, 2001b;Vaisey and Gersho, 1992) and CPAM for image coding and indexing. The image is first segmented into squared homogeneous regions (based on a controllable criterion).…”

“…Each segmented region is then coded into three channels of the CPAM and efficient representation is achieved using vector quantization. Similar to the region and colour co-occurrence matrix of our earlier work (Qiu, 2001b), an effective image descriptor can be derived from the representation stream with little computation-as if the image content descriptor is implicitly ''embedded'' in the compressed stream. We will show that this new image descriptor can be used for effective content-based retrieval and has comparable performances to state of the art methods such as the MPEG7 colour structure descriptor (MPEG7 FCD, 2001) and the colour correlogram method (Huang et al, 1997).…”