SHAPE ENCODING for EDGE MAP IMAGE COMPRESSION Demetrios

SHAPE ENCODING for EDGE MAP IMAGE COMPRESSION Demetrios

SHAPE ENCODING FOR EDGE MAP IMAGE COMPRESSION Demetrios P. Gerogiannis, Christophoros Nikou, Lisimachos P. Kondi Department of Computer Science and Engineering, University of Ioannina, 45110 Ioannina, Greece fdgerogia,cnikou,[email protected] ABSTRACT edges, that is, single binary shape images and bilevel layers of digital maps. DSLS models the edges by digital straight line A method for the efficient encoding and decoding of a shape segments (DSLS) [5]. Compared to standard algorithms like with application to edge map image compression is proposed. JBIG [6], JBIG-2 [7] and MPEG4 CAE [1],[3] DSLSC pro- The method relies on the modeling of the manifold of a shape vides better results, as it fully exploits the information given by line segments. Then, encoding is performed by collect- by the local straightness of the boundary, which is not the case ing the characteristic features of each line segment, namely for the other methods. the starting and ending points and the number of points con- DSLSC is further improved in [8], where the segmenta- tributing to the computation of the corresponding segment. tion of the alpha plane in three layers (binary shape layer, The reconstruction of the shape may be obtained by uniform opaque layer, and intermediate layer) is employed. Exper- sampling of points from each line segment computed in the imental results demonstrated substantial bit rate savings for encoding process. Experimental analysis demonstrates that coding shape and transparency when compared to the tools in case of employing a robust and efficient line segment de- adopted in MPEG4 Part 2. tection algorithm, the proposed encoding/decoding scheme Discrete straight lines were also employed in [9] for shape exhibits high compression rates, compared to widely used encoding and improvement of the compression rate is reached lossless compression methods, while providing low distortion by carrying out a pattern substrings analysis to find high re- values. dundancy in binary shapes. Index Terms— Binary image compression, image repre- A lossless compression of map contours by context tree sentation, line segments, shape encoding, shape reconstruc- modeling of chain codes is described in [10]. An optimal tion. n-ary incomplete context tree is proposed to be used for im- proving the compression rate. 1. INTRODUCTION A JBIG-based approach for encoding contour shapes is introduced in [11], where a method is presented that manages Shape representation is a significant task in image storage to efficiently code maps of transition points, outperforming, and transmission, as it can be used to represent objects at in most cases, differential chain-coding. a lower computational cost, compared to non-encoded rep- In this work, we propose a scheme for efficiently com- resentations. For example, the widely used MPEG4 Part 2 pressing edge map images. The rationale is to model the object-based video standard uses shape coding for describ- manifold of the edge map image by fitting line segments to it. ing regions, called video object planes, that represent an ob- Then, each line segment is encoded by its starting and ending ject [1]. In that case, accurate shape encoding leads to better points and the number of points contributing to its computa- preservation of contour details. tion. The reconstruction of the contour is achieved by uni- The pioneering work in [2], where sequences of line seg- formly sampling points, along the direction of each encoded ments of specified length and direction are represented by line segment. In the remainder of the paper, the compression chain codes was proposed for the description of digitized and decompression algorithms base on line segment fitting are curves, contours and drawings and it was followed by numer- presented along with extensive experimental evaluation of the ous techniques. Shape coding is a field that has been studied method. extensively in the past but it is still very active. Various meth- ods have been studied in [3], including the context-based 2. ENCODING SHAPES arithmetic encoding (CAE), which has been adopted by the MPEG4 Part 2 standard. Line segments are important features in computer vision, as The digital straight line segments coder (DSLSC) was in- they can encode rich information with low complexity. We troduced in [4] for coding bilevel images with locally straight take advantage of this feature for encoding a 2D set of points describing a shape as a collection of line segments that ap- Algorithm 2 Image decompression proximate the manifold of the shape, by assuming that the input: A set of features S that encodes an image I. manifold is locally linear. The initial and ending points of output: The reconstructed image I. each line segment may be considered as the characteristic Recover the characteristic points P = fpi; i = 1 :::Kg, points carrying the compressed information that can repro- based on initial point and transitions encoded in S. duce the initial shape. The larger the number of characteristic for i=2:K do points is, the better shape information is preserved. Produce the set of points R, e.g. [13], containing the A line segment ϵ may be described by its starting and end- points of the line segment from pi−1 to pi. ϵ ϵ ing points xs and xe respectively. The collection of the Set the pixels of I corresponding to coordinates of points starting and ending points of all the segments modeling the in R on. shape manifold are the characteristic points of the shape. Note end for that the characteristic points are ordered. Moreover, since the traversal of the line segments is known, the line segment can be described by its starting point and the transition vector to- 3. EXPERIMENTAL RESULTS wards the ending point. The ending point of one segment is the starting point of its successor in the traversal order. Even- In this section, the experimental investigation of the proposed tually, the shape can be encoded by selecting an arbitrary ini- method is presented regarding its robustness and efficiency. tial point from the characteristic points and by the correspond- To that end, a compression-distortion study was carried out. ing transition vectors after visiting each segment based on the Compression was computed as the ratio of the file size be- traversal order, in a similar manner described in [12]. tween the compressed and the original files. Various lossless methods were considered and the corresponding size of the To reconstruct the image we need to reconstruct all the output files they produced was used as the reference original points contributing to the computation of each line segment ϵ file size. The methods against which we compared the pro- based on the characteristic points. In principle, a line is mod- posed framework are the CCITT G4 standard [14] (denoted eled by the parametric equation Ax + By + C = 0. where as FAX4 herein), adopted amongst others by the TIFF image x; y; A; B; C 2 R and (x; y) is a point laying onto the line. file format for binary images, and the widely used standards If the starting (x ϵ) and ending (x ϵ) points of a line segment s e JBIG [6] and JBIG2 [7]. ϵ are given, determining A, B, C is trivial. Thus, starting from As far as the distortion is concerned, a twofold compu- point x ϵ and following the direction of the line segment with s tation was performed in terms of measuring the loss of in- a predefined step λ 2 R+ each time, we may reconstruct (ap- formation and the similarity between the initial and the final proximate) the initial points. The value of the step λ controls edge map images. Therefore, the distortion index adopted by the density of the result: the higher its value is, the larger is MPEG4 [15], given by the number of extracted points. In case of points laying on an image grid, integer arithmetics need to be considered and se- Number of pixels in error D = ; (1) lecting λ = 1 yields the algorithm of Bresenham [13] which R Number of interior pixels may reconstruct the line segment pixels efficiently and handle the aliasing effect. was also used in this work. The Hausdorff distance between Algorithms 1-2 describe the proposed framework for the original edge map X and the reconstructed edge map Y s, compression/decompression of bi-level images of edge maps. given by DH (X; Y ) = max minfjx − yj1g; (2) Algorithm 1 Image compression x2X y2Y input: An edge map image I, representing shapes. was used to measure the similarity between X and Y . output: A set of features S that encodes image I. As mentioned in section 2, the proposed compression Detect the line segments that describe I. Let K be the num- method uses a line segment fitting algorithm. There are a lot ber of line segments detected. of methods that have been proposed in the related literature. Detect the traversal order of the line segments. A widely used method is the Hough Transform (HT) [16] Refine shape, i.e. close gaps between line segments. Ex- and its variants [17]. However, the principal goal of HT is to tract the characteristic point P = p ; i = 1 :::K, based i detect lines or line segments in a coarse level. Thus, it cannot on the shape traversal. be used for detailed description of shapes. This fact has also S = fp g. 1 been observed for some variants of HT in our previous work for i=2:K do [18]. S = S [ fdx; dx = p − − p g. i 1 i Another method for line segment detection is polygon ap- end for proximation [19]. Having the ordering of the points, line segment computation begins from an arbitrary point, and by traversing the shape, fits a line segment to all the points that by FAX4 [14], 25% of the bits used when compressed by have been visited.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    5 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us