Skip to content

A project report submitted in fulfillment of the requirements for B.Tech. Project

Notifications You must be signed in to change notification settings

shadowshot-x/Generative-Image-Inpainting-using-Deep-Learning

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 

Repository files navigation

Generative-Image-Inpainting-using-Deep-Learning

A project report submitted in partial fulfillment of the requirements for B.Tech. Project

Abstract

Image Inpainting has been one of the most ancient problems in the field of Computer Vision. The Utility of this problem lies from the field of surveillance to creation of competent datasets . In real life applications, the complexity of some computer vision tasks is increased due to some corrupted or missing values of pixels in images. On this scale, it is very difficult to estimate the value of pixels in the missing regions. A lot of models have been proposed to solve the problem of missing pixels and address the problem of large missing values. However in this field, getting satisfactory results is somewhat complex. In this Btech. Project, We have proposed a solution approach that use architectures based on Deep learning that helps to solve this problem. Generative Adversarial Networks are highly supportive and helpful for the major task of image completion. Therefore, in GANs,a trained Least Squares GAN (LSGAN) architecture has been utilized for completion of missing parts of images which recreates the missing portion by generating the image closest to the corrupted image in coarse patches. This is done by subjecting the image generation of LSGAN to Perceptual and Contextual Loss which generates a realistic looking image similar to the data distribution of image. After this a Refinement network is trained on images with noise to remove the noise which dditionally enhances the quality of resultant images utilizing an Auto Encoder Network procedures and hence provide complete and enhanced pictures for computer vision applications. The AutoEncoder Model also helps to generalize some outlier results generated by the Generative Network. This model is inspired by Context Encoders and Progressive Inpainting approach.The Experimental outcomes show that the proposed approach improves the Peak Signal to Noise ratio and Structural Similar ity Index values by 2.5% and 2% than the existing Techniques in use.

Keywords: Auto-Encoder Network, Deep learning, Image Completion,Image Enhancement, Least Squares GAN .

Contents

  1. The Final Research Report
  2. The Final Research Presentation

About

A project report submitted in fulfillment of the requirements for B.Tech. Project

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published