IMAGE ENHANCEMENT THROUGH EDGE PRESERVATION: A MULTI-LEVEL HYBRID DENOISING, RIDGELET AND CURVELET BASED TECHNIQUE

Abstract

Medical image is one of the post power full tools for studying neurological disorders. Epilepsy is one of the them. Over 50 millions peoples are suffered from epilepsy. Medical image can used to study the epilepsy, as the brain stroke, brain tumor, and brain injuries are the most common causes of it. Here image pre-processing plays a significance role to enhance the quality of the input image. Further, image registration, noise removal, image enhancement, etc. are the common processes involved in pre-processing. An Efficient real time embedded solution for medical image enhancement is needed. In the area of machine learning (ML) and artificial intelligent (AI) solution, most of existing algorithms are software based and their hardware implementation is difficult. However, hardware implementation of such algorithm is needed for real time performance. This work proposes a image enhancement technique based on wavelet, Ridgelet and curvelet transform. Here a modified Haar wavelet transform with sliding window-based overlapping execution is use for image denoising with a dedicated embedded hardware implementation. The proposed algorithm provides a solution for image enhancement by image’s edge preservation. Further the image edges have been preserved by Ridgelet and modified curvelet transform based technique. Performance of the projected image enhancement algorithm has been evaluated in terms of many statistics approached namely peak signal-to-noise ratio (PSNR), mean square error (MSE), mutual information (MI), universal image quality index (UIQI), geometric mean (Gm), and gradient (Gd). The results obtained by applying the projected algorithm demonstrate the significant improvement in output image quality. Further, the embedded version of the proposed algorithm has been successfully implemented into ESP32-WiFi Cam module powered by tiny 32-bits processor architecture, small amount of primary memory along with dedicated SD card slot for storage. Further it has been observed that the in-place calculation with software pipelining technique offers memory and time efficient solution. Further, lower computational overhead results in fewer resource requirements which make the proposed solution suitable for real time and portable.