Texture and Structure Based Image Inpainting Techniques-Median Filter Based Technique

Abstract

Image inpainting is an image processing technique to remove unwanted objects from the images by diffusing the pixel information from the neighborhood pixels. Image inpainting techniques are in use over a long time for various applications like removal of objects from the images, scratch removal and so on. The algorithms developed can be broadly categorized into two major classes: Structure inpainting and Texture inpainting. Structures inpainting techniques are used to remove objects or inpaint the image against a homogeneous background where as the texture inpainting techniques are used to diffuse texture map in the inpainted area. Most of the structure inpainting techniques propagate the information from the inpainting boundary into the inpainting area. Number of iterations are to be selected manually prior to the inpainting in order to avoid instability. In this work we present simple and unexplored inpainting technique which achieves better results in terms of both visual quality and analytical quality.  The technique is inherently stable and converges.

In this project we propose three techniques:

• Median Inpaiting based technique.
• Gaussian Pyramidal Technique.
• Local Binary pattern based Image Inpainting Technique.

The proposed algorithm is based on propagating 1) median value, 2) Gaussian filtered values and 3) LBP values of pixels from exterior area to be inpainted into the inner area.  Results show remarkable improvements in terms of PSNR over the methods proposed by Bertalmio and Roth.