S. Inglis and I. H. Witten. Compression-based Template Matching. In J.A. Storer and M. Cohn, editors, Proc. of the  Home/Search   Document Details and Download   Summary   Related Articles   Check

....a match between two symbol bitmaps that actually represent two different characters, we say a substitution error has occurred. For the PM S based JBIG2 system, as we will explain shortly, adopting a pattern matching criterion that is robust against substitution errors is particularly important. In [9], a pattern matching technique is introduced that is robust against character substitutions but very computationally demanding. In the JBIG2 system we built, we adopt the Hamming distance matching criterion. We measure the percentage of different pixels between two symbols (binary patterns) In ....

....image. In real applications, when using PM S for lossy coding, extreme care should be taken in selecting a proper pattern matching criterion so that the system suffers minimal character substitution. 15 Usually this means a more sophicated pattern matching technique that takes longer to compute [9]. However, in our experiments, we use the same Hamming distance matching criterion for both systems. We set the mismatch threshold to be low for the PM S encoder to suffer rare substitution errors. Our experiments are done on a Pentium Pro 200MHz, running Red Hat Linux 6.0, with 64MB physical ....

S. Inglis and I. H. Witten. Compression-based Template Matching. In J.A. Storer and M. Cohn, editors, Proc. of the

....processing. While Howard proposes a methodology for scalable lossy compression, his approach lacks a hierarchical representation suitable for progressive transmission. 5 An important approach to document compression, suggested by Ascher and Nagy [9] and later formalized by Witten et al. [43, 110, 111, 112], is based on the repetitive nature of text components. In a departure from pixel level to symbol level coding, marks found within a document are coded. The method is very similar to patternmatching and substitution algorithms [10, 47, 64, 69, 70, 85, 113] It is based on the measurement of ....

.... matching method such as Pratt s combined symbol matching [85] Holt and Xydeas weighted andnot method [36] Holt s combined size independent strategy [35] and Johnsen s generic pattern matching and substitution [47] Another method of matching, called compression based template matching [43] was suggested by Inglis and Witten and was later improved upon by Zhang and Danskin [116] it was based on mutual information between the component and the prototype. The locations of the components are ordered in natural reading order, indexed with respect to each other (a component is ....

[Article contains additional citation context not shown here]

S. Inglis and I. Witten. Compression-based template matching. In Proceedings of the IEEE Data Compression Conference, pages 106--115, 1994.

....cluster prototypes. After all observations have been processed, the prototype map typically looks like the one shown in Figure 1. A number of component matching algorithms can be used, including simple XOR, weighted XOR, Boolean AND NOT, blur HitMiss [2] and compression based template matching [5, 7, 9]. Note that by clustering we are avoiding the cost of performing OCR, and a perfect clustering is not needed. Clusters of components are represented by a single prototype, and depending on the sample space there exists finite variability in these clusters. For some clusters, the amount of in class ....

S. Inglis and I. Witten. Compression-based template matching. In Proceedings of the IEEE Data Compression Conference, pages 106--115, 1994.

....contains one representative of each pattern class is gradually built up, and the patterns in the library are compared with the patterns to be encoded. The success of the compression depends on the accuracy of pattern matching. A number of pattern matching algorithms were proposed in [4] 5] and [2]. The existing algorithms fall into two main types, depending on whether global or local criteria are employed in obtaining the matching decision. ffl One algorithm based on global criteria is Combined Symbol Matching (CSM) This algorithm computes a weighted count of the error pixels, in which ....

....CSIS looks for four error pixels clustered in a square, and rejects a match if such pattern is found. These rules are heuristic and do not have a firm theoretical basis. Most of the rules are not scalable. ffl Inglis and Witten proposed a compression based pattern matching scheme (CTM) in [2]. CTM produces better results than previous pattern matching methods. However, the context based compression model used by CTM is not a close approximation of the true cross entropy between pairs of patterns. When comparing two patterns one filled with black pixels and one filled with white ....

Inglis,S. and I.H.Witten, "Compression-based template matching," Proc. IEEE Data Compression Conference, pp.106-115, IEEE Computer Society Press, Los Alamitos, CA, 1994.

....is compressed using Moffatt s two level context based method (following [11] The glyph indices are coded using PPMC. The glyph positions are coded using structure based position coding developed in [15] Several recent papers have focussed on how to measure the distance between pairs of bitmaps [7, 14] via cross entropy measures. These measures approximate the number of bits needed to encode one bitmap given the other under various models. In this paper we use the measure proposed in [14] The main contribution of this paper is a new method for the partitioning step. Recent works have used ....

S. Inglis and I.H.Witten, "Compression-based template matching," Proc. IEEE Data Compression Conference, pp.106-115, IEEE Computer Society Press, Los Alamitos, CA, 1994.

....code them. On the other hand, if two different patterns are accepted as matched, the reconstructed image will have different text from the original image. This is unacceptable for most document image compression applications. A number of pattern matching algorithms were proposed in [5] 6] and [2]. Most algorithms are based on analysis of the error map, which is the logical exclusive or of the two images. Existing algorithms fall into two main types, depending on whether global or local criteria are employed in obtaining the matching decision. ffl One algorithm based on global criteria is ....

Inglis,S. and I.H.Witten, "Compression-based template matching," Proc. IEEE Data Compression Conference, pp.106-115, IEEE Computer Society Press, Los Alamitos, CA, 1994.

....the images dominates the size of the coordinates, and coordinate compression is unimportant. ffl When bi level images are too big to be glyphs I use the pattern of bits in Figure 4 to predict bits one by one. This 6 bit pattern (similar patterns appear throughout the literature: see for example [14]) picks up both horizontal and vertical patterns in the input image, providing a kind of cheap 2D compression. I experimented with larger and smaller patterns, but this one seemed to give the best compression on my trace suite. I assign a third value to out of image pixels to account for image ....

....text to characters in the scanned font. This is a much harder problem than the one I have addressed here because of the noise and registration issues associated with scanning. A recent system, which achieved 37.8:1 lossless compression on a library catalog scanned at 400dpi was described in [14]. The high compression ratio is because of all of the white space in a scanned document. HBX s compression ratio is lower because its input has denser content. In this example HBX has achieved a 5.5:1 expansion with respect to the compressed source text. We can see how to get to an expansion of ....

Inglis, Stuart, and Ian H. Witten, "Compression-based template matching," Proceedings 1994 Data Compression Conference, March 29-31 Snowbird Utah pp 106-115.

....template matching method for lossless compression. Unlike the two previous sections, the best result is obtained by the CSIS matching method and the worst by the combined method. These interesting findings are discussed further. 6. 1 Template matching methods Following work by Inglis and Witten [IW94] we divide methods for template matching into two categories: local and global matching. Local methods perform analysis on a small section of the error map (or sometimes the actual image) and return a boolean match or differ result. Global methods examine the whole error map and generate a ....

....components, we use the pairwise compression of components introduced in Chapter 2. A clairvoyant context can be used to estimate the information content of each pixel, because it is not necessary to be able to compress and decompress the component. We use the same contexts as Inglis and Witten [IW94] and these are shown in Figure 6.2(a) and Figure 6.2(b) Larger contexts may be more suitable 102 (a) Conventional (b) Clairvoyant Figure 6.2: Two different types of contexts for larger components, or for higher digitisation resolutions. It is normal to use an adaptive model, accumulating ....

[Article contains additional citation context not shown here]

Stuart J. Inglis and Ian H. Witten. Compression based template matching. In Proceedings of the IEEE Data Compression Conference, Snowbird, Utah, pages 106--115, Los Alamitos, CA, USA, April 1994. IEEE Computer Society Press.

No context found.

Inglis, S. and I. H. Witten, "Compression-based template matching," Proc. IEEE Data Compression Conference, pp.106-115, IEEE Computer Society Press, Los Alamitos, CA, 1994.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC