Vector Quantization is a method in which many data is represented by only one vector, therefore, many kinds of the Applications are proposed. For example, data compression algorithm for color digital images which consist of many data is proposed. In this paper, we propose a new pattern recognition method for binary images using Vector Quantization. We develop an automatic classification of vector patterns and show the experimental results applied to simple binary patterns.
This paper pesents the metods of image sharpening. A system of linear equations is formulated from the point spread function which is approximated with Gauss distribution. Unique image can be obtained by this method since the method satisfies both the necessary and sufficint conditions for obtaining solutions. Computer simuations are made to sharpen the image by the Gauss distribution. Using the practical image, the usefulness of the method is verified. The new operation metod of image sharpening is also proposed. The operation method is equivalent to the method of linear equations. There are some merits in processing time and memory capacity.
A method for improving coding efficiency of coding method using neural network is proposed. In the method, an original image is divided into blocks and according to the characteristic of each block, certain independent neural networks are employed. The experimental results show that by using the method, the image qualities of the reconstructed image are increased.
The effect of noise parameters on TV picture degradation was studied by using a composite noise generator which can control noise parameters such as power, APD, ACR, and occurrence frequency. The disturbance was subjectively evaluated. The results show that the effect of noise power on TV picture was larger than those or APD and ACR, and the evaluation scores were different with backgraund pictures. The experimental equation was obtained to assume the relationship between the noise parameters and the evaluation scores so that the evaluation score can be estimated when the noise parameters were given.
This paper describes a multidimensional color vector quantization method on the CIE-L^*a^*b^* uniform color space, HVC munsell renaotation color spce and YIQ color space. And make the detailed comparison of quntized image qualities.
Capability of the progressive reproduction is a factor desirable for embedded data compression scheme in image database systems. This paper presents an image data compression scheme based on a hierarchical bicubic spline interpolation and vector quantization. In our method, input images are decomposed hierarchically into a decimated image and some decimated residual images with various decimation ratios, e.g. 8 : 1,4 : 1 and 2 : 1. In order to remove redundancy among samples, an adaptive decimation method is introduced. And then the decimated samples are quantised by the tree search vector quantizers. Simulation experiments are done for the three-stage hierarchical decimation/interpolation and the four-dimensional eight-bit tree search vector quantization.