Supervised Saliency Map Driven Segmentation of the Lesions in Dermoscopic Images
Lesion segmentation is the first step in the most automatic melanoma recognition systems. There are some deficiencies and difficulties in dermoscopic images that make the lesion segmentation an intricate task e.g., hair occlusion, presence of dark corners and color charts, indistinct lesion borders, and lesions touching the image boundaries. In order to overcome these problems, we proposed a supervised saliency detection method specially tailored for dermoscopic images based on the discriminative regional feature integration (DRFI) method. DRFI method incorporates multi-level segmentation, regional contrast, property and backgroundness descriptors, and a random forest regressor to create saliency scores for each region in the image. In our improved saliency detection method, mDRFI, we have introduced some features as regional property descriptors and proposed a novel pseudo-background region to boost the performance. The overall segmentation framework uses the saliency map to construct an initial mask of the lesion through thresholding and post-processing operations. The initial mask is then evolving in a level set framework to fit better on the lesion boundaries. Results of evaluation experiments on three public datasets show that our proposed segmentation method outperforms the other conventional state-of-the-art segmentation algorithms and its performance is comparable with the most recent deep convolutional neural networks based approaches.
READ FULL TEXT