The Linear Predictive Identification Of Yaw Stretch Areas In Side Scan Sonar Imagery

Abstract

In this paper we apply 2-dimensional linear predictive texture analysis to the identification of yaw stretched areas of side scan sonar imagery. The image is first divided into blocks, the dimension of which is determined by the correlation distance of the 2-D autocorrelation function. For each block of the training image, an LPC analysis is conducted and, by vector quantization, three types ofcodebook textures are established. ' b o code words are associated with yaw stretch areas and a third is related to the normal background. Each block of the scanning imagery is mapped via least distortion into either a background or a yaw stretch codeword. The yaw stretch blocks are then analyzed for horizontal patterns of texture which are indicative of yaw stretch areas. These areas are then tagged and passed to an image recognition processor so that suitable adjustments to the detection algorithm can be made. Fig. 1. The 2-D Causal Mask Used in Linear Predictive Texture Analysis