Skin cancer is one of the most frequent cancers among human beings. Whereas, malignant melanoma is the most aggressive and deadly type of skin cancer, and its incidence has been quickly increasing over the last years. The detection of the malignant melanoma in its early stages with dermoscopic images reduces the mortality considerably, hence this a crucial issue for the dermatologists. However, their interpretation is time consuming and subjective, even for trained dermatologists. The current computer-aided diagnosis (CAD) systems are mainly noninteractive in nature and their prediction represents just a cue for the dermatologist, as the final decision regarding the likelihood of the presence of a melanoma is left exclusively to him/her. Recently, developing CAD schemes that use image retrieval approach to search for the clinically relevant and visually similar lesions has been attracting research interest. Although preliminary studies have suggested that using retrieval might improve dermatologists' performance and/or increase their confidence in the decision making, this technology is still in the early development stage with lack of benchmark evaluation in ground-truth datasets to compare retrieval effectiveness. A CAD system based on both classification and retrieval would be more effective and robust. This work is focusing on by addressing the various issues related to the development of such an integrated and interactive CAD system by performing automatic lesion segmentation with an adaptive thresholding and region growing approach, extracting invariant features from lesions and classifying and retrieving those using Extreme Learning Machines (ELM) and a similarity fusion approach. Finally, various methods are evaluated with promising results in a benchmark dataset of International Skin Imaging Collaboration (ISIC).