A New Family of Distance Functions for Perceptual Similarity Retrieval of Medical Images

Abstract

A long-standing challenge of content-based image retrieval (CBIR) systems is the definition of a suitable distance function to measure the similarity between images in an application context which complies with the human perception of similarity. In this paper, we present a new family of distance functions, called attribute concurrence influence distances (AID), which serve to retrieve images by similarity. These distances address an important aspect of the psychophysical notion of similarity in comparisons of images: the effect of concurrent variations in the values of different image attributes. The AID functions allow for comparisons of feature vectors by choosing one of two parameterized expressions: one targeting weak attribute concurrence influence and the other for strong concurrence influence. This paper presents the mathematical definition and implementation of the AID family for a two-dimensional feature space and its extension to any dimension. The composition of the AID family with L p distance family is considered to propose a procedure to determine the best distance for a specific application. Experimental results involving several sets of medical images demonstrate that, taking as reference the perception of the specialist in the field (radiologist), the AID functions perform better than the general distance functions commonly used in CBIR.KEY WORDS: Distance function, medical images, content-based image retrieval INTRODUCTION D igital images, which are present in the majority of medical systems, serve to support diagnostic activities. However, for the effective and suitable use of images, these systems must include tools for image management, including fast and effective comparison and retrieval. In the image analysis environment, the ability to compare images automatically is important because the number of stored images is usually very large, precluding a radiologist from making individual comparisons of the images in the entire database. Some groups of radiologists prefer to search in reference libraries, which are composed of sets of typical images. However, the possibility of retrieving images containing patient-related information from specific databases enables the analyst to make more in-depth analyses, for instance, to explore and study the patients from a particular geographic region.The core of content-based image retrieval (CBIR) is the use of intrinsic visual features, which are extracted automatically from the images to describe them while keeping their most relevant characteristics. Such features, usually consisting of numerical values obtained by image-processing algorithms, are used to compare and index images and are usually placed together in a feature vector. Each item of a feature vector is also called an image attribute. CBIR techniques take advantage of index structures that use the similarity of features to speed up the retrieval, doing this automatically. 1,2 In Figure 1, the user places a query, which is done typically by providing the system with an image (query image) ...

A similarity-based data warehousing environment for medical images

“…A feature descriptor is characterized by [9]: (i) a feature extractor algorithm, which tracks down the images, processes their pixel values, produces numeric representations of them, and stores these values in feature vectors; and (ii) a distance function, which produces a similarity measure that is used to determine the dissimilarity between two images based on their feature vectors [10,11].…”

Computer-Aided Diagnosis in Mammography Using Content-Based Image Retrieval Approaches: Current Status and Future Perspectives

“…In the CAD applications, the image content should be represented by a set of extracted image features that meet following two criteria. First, since a highly performed CAD scheme using CBIR approach should not only achieve the high performance in detecting suspicious masses (measured by the area under ROC curve), but also generates smaller “semantic gap” between human vision (the high-level image scene understanding) and computer vision (the low-level pixel based image analysis) [60]. Thus, the CAD scheme should include at least some of the features that are closely correlate with the visually similarity in a multi-feature based CBIR algorithm.…”

“…Other popular texture features derived from the co-occurrence matrices [64] and Fourier transformation [65] have also been tested in developing CBIR schemes. Based on the analysis of medical domain knowledge, one recent study reported that the similarity evaluated based on the combination of four types of image features (color histogram, image texture, Fourier coefficients, and wavelet coefficients) using the feature vector dot product as a distance metric was correlated well with the observed visual similarity [60]. …”

Sistematização da percepção médica na construção de sistemas para recuperação de imagens por conteúdo

“…Muitas medidas de similaridade estão sendo desenvolvidas visando otimizar o processo de recuperação de imagens baseada em conteúdo (Felipe et al 2009). A Tabela 3 apresenta as principais medidas/funções encontradas na literatura (Bugatti 2008), suas fórmulas e um breve resumo descritivo.…”

“…Em contrapartida, há também outros atributos cuja interação é tida como fraca, onde uma grande variação em seu valor não provocaria na imagem uma variação perceptual muito grande. Porém, novamente percebe-se que o principal motivo dessa rejeição para a monotonicidade foi, tal como na transitividade, a substituição "subjetiva" do critério de semelhança durante uma mesma comparação (Felipe et al 2009). …”