University of Texas at Austin
Department of Electrical and Computer Engineering
EE 381K Multidimensional Digital Signal Processing
Instructor: Dr. Brian Evans
Student: Zhou Wang
Sep. 15, 1998
Objective Quality Measurement for Nearly Lossless Image/Video Compression
With the advent of various video compression standards and a proliferation of digital video coding products to follow, it has become increasingly important for the telecommunication, computer and media communities to devise formalized quality/impairment measurements that help to provide effective quality digital video services. Currently, two widely adopted methods are the subjective assessment of Mean Opinion Score (MOS) and the objective measurement of Mean Squared Error (MSE). There are two obvious disadvantages of MOS method. First, the subjective test is very tedious, expensive and cannot be conducted in real time. Second, it is very difficult to be embedded into a practical video processing system because it cannot be implemented automatedly. Although the objective metric MSE is simple and convenient for real application, it is not a good measure for the perceived image and video quality because it cannot reflect the human visual perception of image/vi deo distortions and artifacts. MSE is also not good because the residual image is not uncorrelated additive noise. It contains components of the original image.
This project aims to provide an objective and tractable quality measurement for nearly lossless image/video compression. A nearly lossless image/video signal means the reconstructed image/video after transmission and decoding has very high quality and basically appears no difference with regard to the original signal. However, the data is not completely lossless. That is, there may exist some numerical differences between the original and the reconstructed data. The error pixels are sparsely distributed within the image. The goal of our quality assessment system is to tell the users whether the numerical errors will be noticed by human eyes and how serious the errors are.
The key of the project is to take into account some human visual system features, such as Gamma correction, pattern-color separability, contrast sensitivity function, masking effect, etc. Another important factor is to keep the algorithm simple and fast, so that it can be generalized for video quality assessment. Since Matlab is an easy-use tool for fast implementation of the algorithm, we will use it as a start point. However, the Matlab program cannot achieve a fast speed, which is very important for our implementation, so in later implementations, we plan to modify some part of the time-consuming computations into c functions.
 B. A. Wandell, Fundations of Vision, Sinauer Associates, Inc., 1995.
 P. C. Teo, and D. J. Heeger, "Perceptual image distortion," IEEE International Conference on Image Processing, ICIP'94, pp. 982-986, Austin, 1994.