SOME STUDIES ON ENHANCEMENT OF IMAGES OF HISTORICAL IMPORTANCE

Abstract

Restoration of heritage artifacts such as murals and famous paintings of the past is intended to be used for various purposes. Historical images give us an insight into our culture. It preserves and archives old historical documents, manuscripts and inscriptions like palm leaves. It helps to inform and enrich the general public of our culture and civilization. It assists scholars in their research and adds to cultural development. There is a lot of importance of mural images in our Indian culture. Our ancient temples are full of stories depicted by beautifully carved murals on the walls and ceilings of these temples. The paintings and murals are an easy means of understanding ancient stories at a glance. Digital archive of historical images lays a foundation of preserving, restoring and digitizing images of cultural importance. Murals and sculptures have been an integral part of our culture for thousands of years. A Mural image is any bit of work of art painted or sculpted straightforwardly on a wall, roof, or other perpetual surfaces. Indian Heritage sites are known for their rich content of heritage in various forms; especially India is known for its uniqueness and cultural highlights. They are prone to get deteriorated due to several inevitable reasons like exposure to climate changes or frequent touching by spectators leading to chipping or loss of paint, formation of cracks and color distortions in murals. Hence, restoring these ancient artifacts digitally is a challenging task. Many researchers have conducted studies of historical images in order to preserve and restore them and several techniques are used to enhance degraded historical images. It is found that existing algorithms of image processing are not compatible with historical images due to limited source of the data, complex degradation pattern, ground truth and subjective nature of enhancement. In this thesis, different methodologies are proposed to enhance degraded historical images. Each algorithm is designed in such a way that it overcomes the limitation of existing methods for image enhancement. This thesis work is primarily based on the challenges found in historical images and the proposed mechanisms to overcome these challenges. A thorough study and comparative analysis of several existing techniques for image enhancement and restoration is conducted. The entire study can be categorized as an examination and identification of multiple goals that are divided into three main groups. Each problem is approached by putting up a unique model for its resolution. The major problem with mural images or other historical images is of poor contrast due to poor lighting conditions in which the images might have been captured or due to ageing of images. There is a loss of information in both the cases of under exposed and over exposed vi images. To overcome this challenge two models are designed. Each model efficiently enhances the poor contrast image. While enhancing the contrast of an image, already enhanced region of an image becomes oversaturated and loses the details within an image. Thus, in model 1 a slight washed-out effect was observed which was further improved in model 2 by proposing two different algorithms. Use of fuzzy logics made the algorithms dynamically adaptive in nature. The contrast of overall image is enhanced in such a way that under exposed areas are enhanced with different values of gamma and over exposed areas in the image are enhanced with some different value of gamma. Histogram spread is used to compare the results of contrast enhancement using other state-of-the art methods and proposed methods. It indicates that proposed methods produced more promising results. Existing edge detection methods are complex to implement and fail to produce satisfactory results in case of noisy images. A noise in an image tends to make it look low-quality image or cluttered with probably undesired elements. Concept of Finite State Machine is explored and Cellular Automata is deployed to design an efficient edge detection algorithm which is less complex and less time consuming due to parallel processing. Some transition rules are applied to simultaneously update the values of each cell in Cellular Automata which further provides the edges present in noisy and cluttered images. Cellular computing uses logic gates to perform operations so it is computationally simple and fast due to parallel processing. Partial Differential Equations (PDE) based inpainting methods depend highly upon the surrounding information around the target region to interpolate the values in missing region and are prone to high visual inconsistency. Patch based methods give poor result when it fails to find patch in surrounding to image inpainting region. Target region or the missing region which needs inpainting undergoes the speeded up robust features (SURF) extraction process to describe the features present in the neighborhood of missing area. The proposed method reconstructs the image in a way that looks reasonable to human eye with high accuracy and maximum similarity with ground truth images. The results indicate that the proposed method is efficient enough to interpolate the pixel values in a defected region without depending upon its adjacent pixels for interpolation.