Improved Curvelet Based Self Similarity Methods For Magnetic Resonance Image Processing

Abstract

Medical Imaging techniques are routinely employed to create images of the human system for clinical purposes Multi modality medical Imaging is a widely used technology for diagnosis detection and prediction of various tissue abnormalities This research is focused on development of an improved brain image processing technique for the removal of noise from Magnetic Resonance Image MRI for accurate image restoration Feature selection and extraction in MRI brain images are processed using image registration image fusion and image segmentation The medical images suffer from motion blur and noise for which image denoising is developed through Non Local Means NLM filtering for smoothing and shrinkage rule for sharpening The Peak Signal to Noise Ratio PSNR of improved curvelet based self similarity NLM method is better than discrete wavelet transform with NLM filter In this thesis the improved image registration technique is identified and introduced to align and preserve edges in medical images The improved similarity based Brain Image Registration is employed for the alignment of brain image features from geometric distortion Image registration is performed based on different criteria Depending on the control points registration can be divided as area based methods and feature based methods According to the nature of images they can be classified as global rigid and local non rigid methods In this work similarity feature based automatic rigid image registration method is implemented and its performance is compared with the existing similarity metrics Registration process is validated using similarity metrics Joint Entropy JE and Mutual newline