Abstract
Constructing processes of image transformation manually requires a lot of effort, so several methods to automate it with machine learning, such as neural networks or genetic programming, have been proposed. Most of them are just constructed image filters that calculate an output value from values in local area in each pixel independently. However in several tasks, like area detections, the information of more distant area is helpful to processing. In this paper, we introduce a new neural network model for automatic construction of image transformation. The proposed model is composed of a regular array of the identical evolutionary neural networks, represented Real Valued Flexibly Connected Neural Network (RFCN) we previously proposed, and each RFCN connects with neighbor RFCNs. The proposed model is represented Cellular RFCN (CRFCN). Because of the local connections, each RFCN can consider information of distant area indirectly. We apply CRFCN to three kinds of image transformation tasks comparing with other methods and examine the effectiveness.
