I.J. Image, Graphics and Signal Processing, 2016, 11, 1-9 Published Online November 2016 in MECS (http://www.mecs-press.org/) DOI: 10.5815/ijigsp.2016.11.01 Copyright © 2016 MECS I.J. Image, Graphics and Signal Processing, 2016, 11, 1-9 Development of Algorithm to Reduce Shadow on Digital Image Chok Dong Ooi and Haidi Ibrahim School of Electrical and Electronic Engineering, Engineering Campus, Universiti Sains Malaysia, 14300 Nibong Tebal, Penang, Malaysia. Email: ocdong.ueee12@student.usm.my, haidi_ibrahim@ieee.org Abstract—In this paper, two shadow reduction algorithms have been proposed and implemented using CIE Lab color space. The task of performing shadow reduction is done by executing shadow detection, shadow removal and lastly shadow edge correction in a sequential order. The first proposed algorithm is implemented based on pixel illumination and color information meanwhile the second algorithm is carried out via thresholding of one or more CIE Lab color space channels. The outputs from both proposed algorithms are compared in terms of shadow detection accuracy and required processing period. The proposed methods shown some promising results. Index Terms—Shadow reduction, shadow detection, illumination, color space, channel thresholding. I. INTRODUCTION In object extraction, presence of shadow contributes to falsification of the object‟s shape, inaccurately measured of geometrical properties and also creation of false adjacency between different objects on images [1]. The wrong classification of the shadow as foreground can occur easily since the intensity values of the shadow region are typically differ significantly from the background values. The existence of the shadow can cause objects merging and shape distortion that may result in the failure in object detection procedures or object misclassification [2]. It is possible that the shadow been classified as a part of the foreground objects itself. As a consequence, both objects and the shadow are detected at the same time and therefore being merged together into a single blob. Due to this reason, there is a higher possibility of object under- segmentation to occur. This is because different objects tend to get connected via shadow, which causing both object and shadow got merged together. This greatly affects geometrical properties and appearance of the object to be detected [3]. Shadow can cause inaccuracy in some applications, such as the applications for plant leaves segmentation [4], license plate recognition [5], gait recognition [6,7], analysis of remote-sensing images [8,9], classification of objects from video surveillance system [10], underwater object detection [11] and clustering [12]. Shadow removal presents us a whole different challenge as shadow detection. The target shadow region cannot be simply removed, but has to be effectively replaced with data sampled from the remainder of the image itself. This is done to present new color values of the replaced shadow region that looks reasonable to human eyes. However, the approach to replace shadow region by replicating the texture would require an exhaustive search of appropriate texture patches throughout the image and hence brings up computation load and processing time [13]. This problem worsened upon detection of large-sized shadow region. Filling of large target shadow region is harder due to higher possibility of occurrence of spatial interaction of multiple textures within the region, ending up with mutual influences between several textures. Some restoration algorithms which fill up shadow region via diffusion, could introduces blur that is more observable when it comes to filling of the large target region. These problems, hence limit the capabilities of some algorithms to process shadow region of small size only [14]. Salient features that can be extracted from digital image such as contrasts, sharpness and structure of the color are essential to carry out shadow detection and removal. This project mainly utilizes image information, especially intensity information, with the assumption that the shadow areas are being less illuminated than the surrounding areas [15]. Unfortunately, such information may not be found in a grayscale image. Therefore pre- processing such as RGB approximation and the addition of chrominance and luminance would be needed [16]. Thus, for this project, digital color image would be used as input instead of grayscale image, allowing more effort to be put on development of shadow detection and removal algorithm. As shadow detection and removal covers a wide scope of research, this project aims to address problems of shadow detection and removal given an input of a single still image. This project will use single digital color image as input. The developed algorithms will work without the need of referral image. Furthermore, the development of shadow reduction algorithm does not cover for application on video sequences. Therefore, temporal information will not be considered in this research. Deep learning method for shadow detection and removal, such as the one used by Khan et. al [17], will not be employed in this research. The remainder of the paper is outlined as follows. In Section II, literature review on shadow modelling and CIE Lab color space are presented. The two proposed