|M.Sc Student||Chkhartishvili Julia|
|Subject||Plants Classification by Texture Recognition|
|Department||Department of Quality Assurance and Reliability||Supervisor||Dr. Liviu Singher|
Texture recognition is an important component of pattern recognition and machine-vision systems. The difficulty arises from the vast variations in texture information that are due to changing illumination, imaging conditions, thermal fluctuations, and even rotation and scale. Therefore it may be nearly impossible to devise a unique texture-recognition algorithm that is always successful. However, assuming favorable imaging and other necessary conditions, several texture recognition and segmentation algorithms have been proposed: feature extractions, polygon fitting, and probabilistic relaxation. Each of these approaches has respective merits and drawbacks depending on the type of imagery being processed and, in general, can be computation intensive.
This work introduces a method of generating an optical filter for texture recognition which is invariant to rotation, scale and shift distortions and, at the same time, can recognize particular textures in an image and highlight the corresponding regions. The filters are constructed in real-time in an optical pattern recognition system by an adaptive, iterative numerical approach. The design is formalized as an optimization procedure, for which the filter performance is the function to be maximized. During the training procedure filter parameters are selected to maximize the distinction between the target and other object in the image. The latter problem is solved using the genetic algorithm. Filters obtained in this optimization procedure are binary filters with texture recognition ability. In order to ensure there is no dependency between the simulation result and primary data, the filter design was performed to different entrance images. Protrude advantages of texture recognition filters - complete distinction between different groups of plants and good results with one image in at training set, derive from taking advantage of all the visual information about the object.