Restoration of Images Based on Patch Processing

International Journal of Research and Scientific Innovation (IJRSI) | Volume VI, Issue V, May 2019 | ISSN 2321–2705

Restoration of Images Based on Patch Processing

G Vismaya, Soorej K S

Department of Electronics and Communication Engineering, Jawaharlal College of Engineering and Technology, Mangalam, Lakkidi, Ottapalam, Palakkad, Kerala, India

Abstract – It is observed that there is a great increase in use of images in all fields. The different source has different formats for image and can introduce some kind of distortion in it. The increase in use of images increases the need to develop technology of restoration of clear image from corrupted one. In recent times, image processing methods using local patches has become very common. Patch is generally a small rectangular piece of an image. These are small pixel areas and powerful primitives in image processing. The proposed method implements an image restoration depicting the cases of image denoising and inpainting using patch processing. The work flow is formulated in such a way that for a corrupted image, we may extract patched and reorder them in such a way that they are placed in a shortest path of possibility. Then depending on the quality of the images that is, noisy or containing missing pixels, operations that may smoothen image such as filtering or interpolation is applied to the reordered patches to get a good recovery of the image. PSNR and MSE are analyzed as image quality assessments. Then we may permute the image many times under various iterations ( the number of iterations for a denoising is two and for inpainting is three) and may observe the improvement of peak signal to noise ratio and the decrement of Mean square error as the image quality improvement.

Keywords – Patch based processing, denoising, inpainting, patch based permutations, PSNR, MSE

I. INTRODUCTION

An image worth thousands of words and during the process of acquisition of the image, it may be degraded due to various reasons. The increase in the use of images nowadays increases the need of developing a technology to restore the clear image from the corrupted one. Thus image restoration has become a challenging topic in the image processing field. The primitive methods of image restoration techniques include inverse filtering, wiener filtering, mean filters, constrained least square filtering etc and these techniques has its own disadvantages also. The proposed method implements an image restoration scheme depicting the cases of image denoising and inpainting using patch processing. Instead of processing the whole image at once, we divide the image into patches. The dimensions of the patches will be very low compared to the original image.

Read More