Feature transformation using filter array for automatic construction of image classification

Image classification system with machine learning is significant technique that is important in various fields such as face recognition in biometrics and inspection system in product factory. In this paper, we use the three steps image classification model as following. (1)Image filtering as preprocessing. (2)Extraction of features from the image. (3)Classification using the extracted features. Manual construction of image classification system is hard work because there are enormous various image processing filters and features. Further, proper preprocessing is not only different for each classification problem, but also different for each feature extraction algorithm. In order to solve these problems, we propose automatic construction of image classification system by using genetic algorithm(GA) to select appropriate combination of image processing filters and features in each of different classification problem. In addition, some image processing filters are suitable for parallel processing. We use graphics processing unit(GPU) for parallel processing of image processing filters to reduce computational cost. In our experiment, we confirmed that constructions of image classification system were finished in practical time.