期刊名称:International Journal of Advances in Engineering and Management
电子版ISSN:2395-5252
出版年度:2020
卷号:2
期号:11
页码:567-570
DOI:10.35629/5252-0211432435
语种:English
出版社:IJAEM JOURNAL
摘要:Image out painting is a technique used in the recursive painting of images. It generates a new image that aims to restore the missing and damaged part of the image. It is a widely used computer vision algorithm in deep learning. This technique extrapolates the image without distorting the pixels of the image. This technique is widely used in various novel and existing applications for example- for panorama creation, vertically filmed video expansion. Our project focuses on building a three-phase Generative Adversarial Network (GANs) that enhances the quality of the image. The output image appears realistic to the human eye. The image generated after recursive painting has enhanced quality and high pixel resolution as compared to the input image. The model is trained on 36,500 images of baseline width is 128× 128 RGB encodings. The context encoders used in this model adversarially draw the missing region of the image from the surrounding image pixels. This model uses local discriminators combined with the global discriminator and produces very convincing results. Various masking and preprocessing techniques are used to decrease the loss during processing.