Degree Name

Master of Engineering (Hons.)


Department of Electrical and Computer Engineering


Texture is a surface property of an object which humans routinely use in recognition and classification tasks. In image processing and pattern recognition texture is recognised as one of the main features useful for analysis in a variety of applications ranging from medical imaging to remote sensing. This thesis addresses the problem of texture-based identification of objects by a guided mobile robot. In the situation envisaged the acquisition of the texture image is allowed to take place at any distance. Since the viewing distance of the scene varies, the scale of the acquired image will also vary. The possibility of incorporating scale change arising from varying viewing distance into a multi-scale texture recognition scheme is explored in this thesis. A model akin to the human visual system model in which there is a pre-attentive stage of segmenting the region of interest, followed by an attentive stage of recognition is proposed. Gabor filter, which is known to possess similar characteristic as the visual cortex, is employed to perform the pre-attentive task, i.e. to detect the region of interest. In order to reduce the computation time of the pre-attentive stage, a composite Gabor filter is introduced. In the attentive stage, a multi-stage, hierarchical classification technique using Gabor filter is used. The performance of this method in terms of the classification accuracy is then compared with other statistical texture classification techniques. Two most widely used texture classification techniques are chosen^ namely the gray-level cooccurence and the frequency domain based methods.