Software tool for contrast enhancement and segmentation of melanoma images based on human perception Irene Fondón 1 , Qaisar Abbas 2,3 , M. Emre Celebi 4 , Waqar Ahmad 2,3 and Qaisar Mushtaq 2,3 1 Department of Signal Theory and Communications, School of Engineering, C\ Camino de los Descubrimientos s/n 41092, Seville. Spain 2 Department of Computer Science, National Textile University, Faisalabad, Pakistan. 3 Center for Biomedical imaging and Bioinformatics, Key Laboratory of Image Processing, Faisalabad, Pakistan. 4 Department of Computer Science, Louisiana State University, Shreveport, Louisiana, USA. irenef@us.es , drqaisar@ntu.edu.pk, ecelebi@lsus.edu, waqar@ntu.edu.pk, qaisarmushtaq@ntu.edu.pk Abstra c t . In this paper we present a software tool for melanoma border detection ( MBD). It has been designed to be incorporated in any Computer Aided Diagnosis Tool ( CAD) for early detection of melanoma in mass screening programs. The tool is completely automatic, posses a user-friendly interface and does not require any specific hardware. The main steps followed by the implemented algorithm are: uneven illumination correction, color contrast improvement and color image segmentation. All of them are performed in the uniform color space CIE L * a * b * in order to achieve a complete adaptation to human color perception. The program is able to provide not only the final obtained segmentation result but also intermediate graphical outcomes, guiding the user in the process of melanoma detection. This simple, friendly but powerful interface can serve as a support for the medical personnel in the melanoma diagnostic process. The MBD software and some samples of the dermoscopy images used can be downloaded at http://cs.ntu. edu.pk/research.php. K e ywords. Software tool, skin cancer, melanoma border detection, dermoscopy, contrast enhancement, hill-climbing 1 Introdu c tion Malignant melanoma is the most dangerous type of skin cancer [1]. However, when early detected, the survival rate of patients increases about 99 percent [2]. In this sense digital dermoscopy, that is, a skin imaging technique widely used for pigmented skin lesion inspection, constitutes a powerful tool [3] mainly because its non-invasive nature and its capability of RSWLFDO PDJQLｿFDWLRQ WKDW reduces WKH VXUIDFH UHﾀHFWLRQ and makes the subsurface structures more visible [4]. However, even for an experienced dermatologist, an accurate classLｿFDWLRQ rate of skin lesions based only in the use of digital dermoscopy is less than 90% [5]. Stated this, a computer diagnosis ( CAD) tool is highly recommended and adopted to improve skin lesion FODVVLｿFDWLRQ accuracy [6]. These computerized tools usually begin with the isolation of the desired region, cancerous skin, from the background, usually healthy skin. An effective segmentation can dictate the eventual success of the analysis. However, melanoma border detection (MBD) is a difficult task due to low contrast and color variations [7]. A variety of techniques have been proposed in the past for skin lesion detection or segmentation [8]ｱ[11] mainly based on three well-known techniques: thresholding, clustering, and region growing. Most of these lesion area detection methods are limited to grayscale image processing or non- uniform color spaces. They are not designed to effectively deal with the important color information present in the images. To overcome this problem a new technique was proposed in [12]. In this article, a novel automatic MBD technique was developed based on a perceptually oriented color space, i.e. CIE L * a * b * to achieve a complete adaptation to human perception of color. The MBD technique followed three steps. In the first one, the original RGB color image is transformed to the uniform color space CIE L * a * b * . Afterwards, a contrast improvement algorithm was applied adjusting and remapping the LQWHQVLW\ YDOXHV RI WKH SL[HOV LQ WKH VSHFLｿHG range. The second stage is the color segmentation by the Hill-Climbing technique [13]. As a result a labeled image is obtained where the region of interest (ROI) is detected. The last step, consist on the isolation of this ROI applying the Otsu thresholding technique and providing the 45