A Novel Clustering-Based Image Inpainting Model Using the Loermgan Algorithm
Ishan Sharma, Yongwei Nie · 2022
The process of recovering the damaged or missing areas from an image utilizing information as of known portions is termed Image Inpainting. To refurbish the damaged image into a new one alike an actual image, numerous sophisticated methodologies have been established to date. Nevertheless, in the case of images with a larger missing region, these models are not effective in addressing the problem. Likewise, these methodologies are ineffective towards the edge. Therefore, by utilizing the Log of Exponent Rule Generative Adversarial Network algorithm, a novel clustering-centric image inpainting system has been proposed here. Initially, two significant steps, namely (i) noise removal, and (ii) Contrast Enhancement (CE), are performed to pre-process the input images. After that, by utilizing the Adaptive Max One-Sided Box Filter (AMOSBF) algorithm, the pre-processed images’ edges are well-preserved. Then, the most needed features are extracted as of the edge preserved images. Next, by employing Supremum Distance Fast Density Peaks Clustering Algorithm (SDFDPCA), the features being extracted are clustered. Next, the proposed model, termed Log of Exponent Rule Mish Generative Adversarial Network (LOERMGAN), which reconstructs the actual images effectively, is fed with the clustered features and also the masked image and the mask itself. In this research, the openly accessible datasets termed ADE20k, Paris, and Places2 are utilized. Subsequently, the outcomes obtained are analogized with the prevailing methodologies. The experiential outcomes displayed that the proposed model outshines the other prevailing methodologies by effectively reconstructing images.