Restoration of Noisy and Noiseless Fence Occlusion Images
- PDF / 6,475,611 Bytes
- 12 Pages / 612 x 792 pts (letter) Page_size
- 72 Downloads / 157 Views
Restoration of Noisy and Noiseless Fence Occlusion Images M. Varalakshmammaa,* and T. Venkateswarlua,** a
Department of ECE, S.V University College of Engineering, Tirupati, A.P., 517502 India * e-mail: [email protected] ** e-mail: [email protected]
Abstract—A new approach is presented to restore the image from noisy fence images. When people capture the images at Zoos, parks and gardens the fence affects the authentic appeal of the object behind it. Due to the acquisition channels, the noise will be added to the images. So, removal of the fence and noise in these images is necessary to improve the appearance of the desired objects. Segmentation of the fence from the noisy image is very difficult because these are extended into the entire image region. In this paper, segmentation of the fence is done in both noiseless and Gaussian noise corrupted images. Segmentation of the fence is achieved using a graph cut technique. Morphological operations are applied to improve the fence mask. Removal of the fence is done with a hybrid inpainting technique. From the De-fenced image, noise is removed using Conventional Neural Networks. Qualitative and quantitative results show the effectiveness of the proposed approach. Keywords: segmentation, denoising, dilation, de-fencing, inpainting, Conventional Neural Networks, graph cut DOI: 10.1134/S1054661820030281
1. INTRODUCTION Images are often corrupted in the acquisition channel or editing. The restoration technique aims are to get a visually pleasing image. Image denoising and inpainting are two restoration techniques useful itself and are an important preprocessing step in many applications. Denoising of an image is necessary when an image is corrupted by additive white Gaussian noise, which occurs normally in image acquisition channels. Inpainting is done when some pixel values in the image are missing or people want to remove more sophisticated patterns (like text, fence or other objects) from the image. Image de-fencing is a real-world problem in the inpainting. Restoration of the image from the fence patterns is called as image de-fencing [1]. Tourists capture the photographs in the historical places, zoos and monuments at these places fences and barricades are providing for the protection. The photographs taken at these places contain the fence. These fences are unavoidable and this will spoil the authentic apple of the image captured. So this needs to be removed for the better quality of the image. This is done by the image de-fencing algorithm. The image de-fencing has two problems. The first one is a segmentation of the fence and the second one is the restoration of the image.
Received January 30, 2020; revised January 30, 2020; accepted February 23, 2020
In the past few years, a lot of research work is going on image de-fencing. Yang et al. [2] to detect the primary fence mask, the color-based classifier trained with samples obtained by clustering the image into super pixels. The final shape is obtained with multiRANSAC and moving least squares. The occlusion
Data Loading...
