Image Compression and Reconstruction Using Feature Extraction
Proceedings of the Fifth International Joint Conference on Artificial Intelligence. Vision. Pages 658--1.
© 1977 International Joint Conferences on Artificial Intelligence
Abstract
- Burr, D. J. and R. T. Chien, "A System for Stereo Computer Vision with Geometric Models," in Proc. 5th International Joint Conference on Artificial Intelligence, MIT, Cambridge, Massachusetts, August, 1977. IJCAI
- Chien, R. T. and L. J. Peterson, "Image Compression with Feature Extraction and Reconstruction," in Proc. IEEE-CS Workshop on Picture Data Description and Management, University of Illinois at Chicago Circle, April, 1977.
- Claire, E. J., "Bandwidth Reduction in Image Transmission," in Conf. Rec., 1972 IEEE Int. Conf. Communications, pp. 39-8 -39-15.
- Duda, R. 0. and P. E. Hart, Pattern Classifi- cation and Scene Analysis, John Wiley and Sons, New York, 1973.
- Graham, D. N., "Image Transmission by Two- Dimensional Contour Coding," in Proc. IEEE, Vol. 55, No. 3, pp. 336-346, March, 1967. 1/ision-6: r.hien
References were extracted automatically from the PDF and may contain errors.
Full text extracted automatically from the PDF; figures, tables and equations may differ. The PDF is the version of record.
IMAGE COMPRESSION AND RECONSTRUCTION USING FEATURE EXTRACTION*
R. T. Chien and L. J. Peterson Coordinated Science Laboratory University of Illinois at Urbana-Champaign Urbana, Illinois 61801
Considerable research has been devoted to the development of methods for reducing the transmis sion bandwidth of video signals such that the re constructed image is of a high quality. Typically, these methods perform a statistical encoding of the intensity levels of a quantized picture, or they perform some transformation and subsequent data reduction in order to achieve bandwidth reduc tion [3]. Such methods rarely achieve data com pression ratios in excess of ten-to-one without noticeable loss of quality in the reconstructed picture.
In attempting bandwidth reductions of video signals, very little emphasis has been placed on the actual contents of the picture. Although methods have been investigated for performing data compression by coding features extracted from digitized pictures, they have not been very successful because of the fidelity criterion [5]. These methods have produced in some instances compressions of at best twenty-to-one. We feel that methods which extract and encode features from the digitized pictures can achieve compression ratios in excess of 100:1. Furthermore, we feel that there are areas of applications in which the contents of the picture is more important than its esthetic quality. In order to determine how prac tical this viewpoint is, we have performed pre liminary experiments on video data compression which encodes features extracted from digitized images.
Edge Finding and Encoding
The pictures we compress are quantized to 64 levels. The edge finding algorithm used was de signed to perform thinning as it finds the edge points so that the line follower has single pixcel lines to follow [1]. In addition, the algorithm averages out much of the influence of the two levels of quantization noise inherent in the vision system. The edge follower attempts to make as long a line as possible by finding edge points which continue in the general direction of a segment up to a distance of three pixcels.
The encoding of these edge segments is achiev ed by approximating them with straight line seg ments [4]. Data compression is determined by calculating the ratio of the number of bits in the intensity array and the number of bits in the encoding. Each point in the encoding needs 17 bits to represent it: 8 bits for each of the x and y directions and a bit to indicate whether the point is an endpoint of its curve segment.
* This work was supported by the Joint Services Electronics Program (U.S. Army, U.S. Navy, and U.S. Air Force) under Contract DAAB-07-72-C-0259.
Results
Using aerial photographs of the Champaign County countryside pictures were compressed and then reconstructed on a Tektronix graphics display terminal [2]. The photographs were character ized by a preponderance of constant intensities with many long boundaries between regions and many long, straight and curving lines. Ratios in ex cess of 100:1 were achieved and the reconstructed pictures appeared as line drawings of the bound aries of fields, highways and streams. Man-made objects were the easiest things to recognize because they . generally have smooth-curved out lines.
Our results are very encouraging not only because we were able to achieve the ratios hoped for, but also because there is considerable room to improve our system. In particular the edge following routine can be changed from its present upward and left bias to that of following a curve completely in the direction it travels, even when it intersects another curve segment. Secondly, the encoding algorithm can be improved in at least two areas. It can use second order approxima tions to the curves and further encoding can be performed on the output of curve fitting routines. Thirdly, better edge finding routinP-s can be found which achieve better segmentations of the picture thereby resulting in better reconstructions of the picture.
@inproceedings{ijcai1977p157,
title = {Image Compression and Reconstruction Using Feature Extraction},
author = {Chien, Robert T. and Peterson, L. J.},
booktitle = {Proceedings of the Fifth International Joint Conference on
Artificial Intelligence, {IJCAI-77}},
publisher = {International Joint Conferences on Artificial Intelligence},
editor = {Raj Reddy},
pages = {658---1},
year = {1977},
note = {Vision},
url = {https://ijcai.org/proceedings/1977/157},
}
TY - CPAPER TI - Image Compression and Reconstruction Using Feature Extraction AU - Chien, Robert T. AU - Peterson, L. J. T2 - Proceedings of the Fifth International Joint Conference on Artificial Intelligence, IJCAI-77 PB - International Joint Conferences on Artificial Intelligence PY - 1977 SP - 658 EP - -1 UR - https://ijcai.org/proceedings/1977/157 ER -Copied
Machine-readable versions of this paper, for tools and AI agents:
Welcome
Future Conferences
AI Journal Division
IJCAI Proceedings
About IJCAI
IJCAI Awards