Image Registration Using Wavelet Transform

Image Registration is an elementary step in many image processing tasks such as medical image analysis, remote sensing, pattern recognition, face detection, motion tracking etc. In this paper, wavelet transform is used to extract features from image. The extracted features are then used to train Neural Network. Affine transformation parameters are used to evaluate the similarity measure between the reference and sensed image sets. Neural Network produces registration parameters (translation, rotation and scaling) with respect to reference and sensed image. Mean Absolute Registration Error and Mutual Information is used as evaluation parameters. Proposed technique for image registration is accurate and robust for distorted and noisy images than using DCT and FFT for image registration as shown by experimental results.