An exemplar-based approach to search-assisted computer-aided diagnosis of pigmented skin lesions .

摘要

Over the years, exemplar-based methods have yielded significant improvements over their model-based counterparts in image synthesis applications. Notably, texture synthesis algorithms using an exemplar-based approach have shown success where traditional stochastic methods failed. As an illustrative example, I present an exemplar-based approach that yields substantial benefits for user-guided terrain synthesis using Digital Elevation Models (DEMs). This success is realized through exploitation of structural properties of natural terrain. Recently, the development of algorithms and data structures that facilitate search in high-dimensional space has led to a proliferation of exemplar-based methods that synthesize image contents and style by tapping into a large collection of images. In addition, as annotated image datasets become increasingly available, the exemplar-based approach is gaining in popularity for image analysis applications.;This thesis addresses the intersection between exemplar-based analysis and the problem of content-based image retrieval (CBIR). A basic problem in CBIR is the process by which the search criteria are refined by the user through the manipulation of returned exemplars. Exemplar-based analysis is particularly well-suited to query refinement due to its interpretability and the ease with which it can be incorporated into an interactive system. I investigate this connection in the domain of Computer-Assisted Diagnosis (CAD) of dermatological images. I will use the analysis of dermatological images using CBIR as context to demonstrate that exemplar-based methods can also yield significant benefits for image analysis if analogous structural properties can be identified. I will present an exemplar-based algorithm for segmenting pigmented skin lesions in dermoscopy images. In addition, I will present a generalized representation of dermoscopic feature for detection and matching. This representation enables us to realize interactive region of interest (ROI) retrieval capability, including a relevance feedback mechanism to facilitate more flexible query-by-example analysis. Finally, I will assess the role of the benefit of this CBIR-CAD approach through both quantitative evaluations and user studies.