Medical Image Registration Based on Feed-forward Neural Networks and Directional Discrete Cosine Transform

Abstract

Registration is used to align two or more pictures taken, for example, at different times, from different sensors, or from different viewpoints. All large systems which evaluate images require the registration of images, which considered as an intermediate step. Examples of systems where image registration is a significant component include change detection using multiple images acquired at different times, fusion of image data from multiple sensor types, environmental monitoring, image mosaicing, weather forecasting, creating super resolution images, integrating information into Geographic Information Systems (GIS) and medical image analysis. In this research, the concentration is on registration of medical images. Analysis of multi-temporal medical images requires proper geometric alignment of the image to compare corresponding regions in each image volume. The proposed image registration technique consists of two steps. In the first step the feature extraction method is performed by using Directional Discrete Cosine Transform (DDCT). Finally the feed forward neural networks based image registration technique is performed .Comparison study in registration process between Directional Discrete Cosine Transform (DDCT) and Traditional Discrete Cosine Transform (DCT) is performed. Experiments have shown that the proposed method provides more accurate results for registration process when using DDCT than traditional DCT.