Optical Wavelet Transform And Its Applications In Image Processing

The development of Wavelet Transform (WT) and its fundamental theory are systematically described in this dissertation. The properties, implementations, and applications of the Optical Wavelet Transform (OWT) are analysed and discussed in detail. The techniques for realization of OWT and applications in image feature extraction and texture segmentation are explored thoroughly. An optical system for two-dimensional Haar WT is constructed by using an angular multiplexing volume hologram technique. Different local features of binary image are extracted simultaneously by using the multiplexing Haar wavelets. The Roberts operator is proposed to be used as wavelet. Roberts wavelet can overcome the limitations of Haar wavelet and can be used for extracting any edges of different orientations of an input image because it is rotational anti-symmetricity. Simula- tions and experiments are performed to confirm the validity. A joint wavelet-transform correlator (JWTC) is proposed. The difficulties generally encountered for the implementation in the frequency domain of OWT can be overcomed by the proposed JWTC. A spatial-division-multiplexing JWTC is al- so designed, which can analyze an image with multi-scale and multi-wavelet, ex- tract and synthesize different features of an input image simultaneously. Diffraction is treated as wavelet transform in a broad sense. The wavelet con- cept of the wavelet theory is extended to non-dandpass family of functions. The Huygens spherical wavelet and the impulse response of Fresnel diffraction are re- written in the form of wavelet. The diffraction distance performs the dilation factor of wavelet theory. Then, a new technique for texture segmentation is proposed based on the Talbot effect of the periodic object. Theoretical analysis and primitive experimental results verify the feasibility of spatial distributed image processing techniques. A scheme for texture segmentation with multi-channel OWT is proposed. This scheme is based on the multi-channel filtering mechanism of human vision in- formation processing theory. The OWT, neural network (NN) and fuzzy c-mean clustering algorithm (FCM), are studied and synthesised to form an intelligent liii- age processing technique to some extent. Based on this approach, a hybrid texture segmentation experimental system is designed and implemented, where the OWT is used for realizing the fast image transform. The NN and the FCM are used for im- age feature recognition. The applications of this system in the segmentation of tex- ture of the typical texture image, rock surface, and bird-view photograph of a city, etc are studied. Fine results of experiments and computer simulations are provided. In order to make the system more compact and robust, we proposed and fabri- cated a novel binary optical element named "optical retina" for the reason that it has the functions of filtering, beam-splitting, and imaging, etc simultaneously. The OWT of an input image with multi-scale and multi-orientation can be obtained in one operation with this element.