Enhancement of Low-illuminance Image Using the illuminant Map Estimation in the Dichromatic Reflection Model

Abstract

Image enhancement is crucial in image processing. There are several reasons for the image possession of poor qualities in an image or video(e.g. unfavorable environmental condition, lack of operator expertise, etc.) The above mentioned factors engender the under-utilization of the offered dynamic range. Therefore, information of an image or a video will be ambiguous and washed-out. The purpose of the image enhancement is to eliminate theses problems. Tone mapping method is one of the necessary processes in order to acquire high quality images in the field of image enhancement. The method is also used to adjust the dynamic range according to the characteristics of the devise. It is further used as a technique to improve the degraded image for its backlight, and the failure in adjusting the exposure. The tone mapping method is generally categorized into two different approaches: global and local. Global approach is to equally apply to all pixels, such as exponent function, logarithm function, and sigmoid function. Local approaches calculate a local mapping from the neighborhood of each pixel, and applies the mapping to a block centered in each pixel. The thesis proposes a method for a single image enhancement, which is affected by low-illuminance condition. The proposed method consists of two steps. First step include the estimation of illumination map, which is normalized 0 to 1. The second step starts with the production of image enhancement via using estimated illumination map from first step based on acquiring image model for a low-illuminance condition. The performance of the proposed method is compared with 6 different algorithms. The experiment tests the image affected by various low-illuminance condition from INRIA holidays dataset. Parameters of each algorithm have been set as the recommendations of each paper follow. The parameter of the weighting distribution of AGCWD has been set to 1.5. The gamma of BEASF have been set to 1.5 as well. The proposed method has demonstrated the production of more logical results than the previous methods have established. The proposed method may also improve the user’s recognition rate in using automatic driving, game, and image processing tools. The thesis expects that the framework of the proposed method would support the study of image enhancement and an image processing.