Texture information processing system with binary optical wavelet element

ABSTRACT The implementation strategy of optical wavelet transform for texture information processing is discussed in this paper.An opto-electronic hybrid system is constructed for texture segmentation, which is based on the multi-channel filteringframework in the early stages of human visual theory. First, a traditional optical system with a Dammann grating as thebeam splitter and a bank of Gabor wavelets as the channel filters is set up for feature extraction, and several clustering algorithms are then used for feature integration. Furthermore, a novel binary optical element with the functions of splitting, filtering and imaging is designed and fabricated to simpliQr the traditional system. The experimental results andthe primary applications are also provided.Keywords: optical wavelet transform, texture segmentation, binary optical element, multi-channel filtering technique 1. INTRODUCTION As one ofbasic features on the visual surface, Texture plays an important role in many image information analysis andapplications. The shape, direction and depth etc. of an object can be furnished by texture. Usually, differences in graylevel or in color among neighborhoods are insufficient for image segmentation. Texture, which includes the spatialarrangement and the statistical information of gray values is needed. Hence, texture segmentation is generally known as adivision of an image into different regions characterized by similar texture.As the main feature in image segmentation, texture is more useful when there are few differences between the objectand the background. On the other hand, the accuracy of segmentation can be improved with texture feature considering.Feature extraction and integration are often two steps of texture segmentation algorithms proposed recently. The formerstep is to measure textures locally with a variable window so that textures can be replaced by feature vectors, multi-resolution is needed for a high accuracy. The development in wavelet theories is fit for texture analysis. With window sizechanged according to frequency, it has optimal joint resolution in both the spatial and spatial-frequency domains. In thelatter step, a clustering algorithm (supervised or unsupervised) is used to integrate the texture features and produce asegmentation.In this paper, the implementation strategy of optical wavelet transform for texture information processing is discussed.An opto-electronic hybrid system is set up, which is based on the multi-channel filtering framework in the early stages ofhuman visual system. A novel binary optical element with the functions of splitting, filtering and imaging is designed andfabricated to simplif' the hybrid system. The experimental results and primary applications are also proposed.