Performance Analysis of Image Inpainting using K- Nearest Neighbor
Manjunath R. Hudagi, Shridevi Soma, Rajkumar Laxmikanth Biradar · 2022 4th International Conference on Smart Systems and Inventive Technology (ICSSIT) · 2022
Image inpainting is a method that focuses to fill the disappeared regions to restore the images back to provide a complete meaningful image. However, the classical methods still struggle with solving the data loss problem because of object and its background. This research analyzes the performance of inpainting model using K-Nearest Neighbor (KNN)-based search method with Bhattacharya distance and the patch extraction is performed using the concept of Bhattacharya distance. Here, the analysis is done by varying the patch size in KNN-based search model for the considered image depending on the percentage of scratch. Moreover, the performance is estimated by considering metrics like Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Second-Derivative-like Measure of Enhancement (SDME), Universal image Quality Index (UQI), Multi-Scale Structural Similarity (MS-SSIM) and Mean Square Error (MSE). From the analysis, it is clear that the KNN + Bhattacharya distance provides effective performance with respect to evaluation metrics and also shows higher efficiency when compared with any other traditional approaches. When the percentage of scratch is 100%, the MSE attained by KNN + Bhattacharya distance with patch size 20x20 is 0.190, MS- SSIM is 0.925, PSNR is 29.01, SDME is 69.02, SSIM is 0.914, UQI is 0.190 for dataset 3.