Image Processing and CGP

Computerized image processing has been studied intensively for many years [32]. Because of the inherent complexity of the problem, stochastic optimization algorithms, including evolutionary algorithms, have been applied to improve existing image processing methods and develop new algorithms [4]. Section 6.2 deals with the automatic design of low-level image filters, ones comparable in quality to conventional filters. Moreover, when the evolved filters are constructed as single-purpose circuits in a field-programmable gate array (FPGA), the implementation cost is usually lower than the cost of conventional solutions. We will describe the basic approach to filter evolution and its extensions, such as the use of a bank of evolved filters, and filtering using extended kernels. Evolutionary design of more advanced image operators such as dilation/erosion filters is presented in Sect. 6.3. As these filters are composed of relatively complex elementary functions (sine, square root etc.), they are primarily intended for advanced image processing software tools. An image classification task will be described in Sect. 6.4, where CGP graphs are used to define transformations in the case of a medical image problem. We will show that CGP image transformations may be used to significantly improve classification accuracy with respect to a collection of predefined image features.