Information Retrieval

Only available on StudyMode
  • Topic: Image processing, Color space, Information retrieval
  • Pages : 11 (3389 words )
  • Download(s) : 71
  • Published : May 11, 2013
Open Document
Text Preview
A Query-by-Example Content-Based Image Retrieval System of Non-Melanoma Skin Lesions Lucia Ballerini1 , Xiang Li1 , Robert B. Fisher1 , and Jonathan Rees2 School of Informatics, University of Edinburgh, UK x.li-29@sms.ed.ac.uk, lucia.ballerini@ed.ac.uk, rbf@inf.ed.ac.uk 2 Dermatology, University of Edinburgh, UK jonathan.rees@ed.ac.uk 1

Abstract. This paper proposes a content-based image retrieval system for skin lesion images as a diagnostic aid. The aim is to support decision making by retrieving and displaying relevant past cases visually similar to the one under examination. Skin lesions of five common classes, including two non-melanoma cancer types are used. Colour and texture features are extracted from lesions. Feature selection is achieved by optimising a similarity matching function. Experiments on our database of 208 images are performed and results evaluated.

1

Introduction

Research in content-based image retrieval (CBIR) today is an extremely active discipline. There are already review articles containing references to a large number of systems and description of the technology implemented [1, 2]. A more recent review [3] reports a tremendous growth in publications on this topic. Applications of CBIR systems to medical domains already exist [4], although most of the systems currently available are based on radiological images. Most of the work in dermatology has focused on skin cancer detection. Different techniques for segmentation, feature extraction and classification have been reported by several authors. Concerning segmentation, Celebi et al. [5] presented a systematic overview of recent border detection methods: clustering followed by active contours are the most popular. Numerous features have been extracted from skin images, including shape, colour, texture and border properties [6–8]. Classification methods range from discriminant analysis to neural networks and support vector machines [9–11]. These methods are mainly developed for images acquired by epiluminescence microscopy (ELM or dermoscopy) and they focus on melanoma, which is actually a rather rare, but quite dangerous, condition whereas other skin cancers are much more common. To our knowledge, there are few CBIR systems in dermatology. Chung et al. [12] created a skin cancer database. Users can query the database by feature attribute values (shape and texture), or by synthesised image colours. It does not include a query-by-example method, as do most common CBIR systems. The report concentrates on the description of the web-based browsing and data

mining. However, nothing is said about database details (number, lesion types, acquisition technique), nor about the performance of the retrieval system. Celebi et al. [13] developed a system for retrieving skin lesion images based on shape similarity. The novelty of that system is the incorporation of human perception models in the similarity function. Results on 184 skin lesion images show significant agreement between computer assessment and human perception. However, they only focus on silhouette shape similarity and do not include many features (colour and texture) described in other papers by the same authors [11]. Rahman et al. [14] presented a CBIR system for dermatoscopic images. Their approach include image processing, segmentation, feature extraction (colour and textures) and similarity matching. Experiments on 358 images of pigmented skin lesions from three categories (benign, dysplastic nevi and melanoma) are performed. A quantitative evaluation based on the precision curve shows the effectiveness of their system to retrieve visually similar lesions (average precision 60%). Dorileo et al. [15] presented a CBIR system for wound images (necrotic tissue, fibrin, granulation and mixed tissue). Features based on histogram and multispectral co-occurrence matrices are used to retrieve similar images. The performance is evaluated based on measurements of precision ( 50%) on a database...
tracking img