Dermoscopic image analysis using pattern recognition techniques from region of interest (ROI) for detection of melanoma

摘要

Patten recognition techniques are widely used for image processing in medical imaging. It provides assistance to physicians and scientists in large scale diagnosis. In this paper, we have proposed an automated system for detecting melanoma from dermoscopic images. We detected melanoma by extracting information from region of interest (ROI) rather than the whole image composed of lesion and background. ROI of image is extracted out by using one of the segmentation techniques named as adaptive thresholding with morphological operators. After extracting ROI, we analyzed the classification results by using texture and color features on the extracted region and compared the results. The images were obtained from one of the hospitals of Portugal utilizing dermoscopy as an imaging methodology. Our extracted algorithm shows better performance than conventional thresholding and based on experimental results it can be concluded that the color features outperform the classification results but it gives more accurate results by combining color and texture features extracted from ROI.