Dan Burdescu

Semantic Based Image Retrieval Using Relevance Feedback

In this paper, we propose a method for image categorization and retrieval, by integrating knowledge from low-level and semantic features extracted from images. The low -level descriptors, like color, position, dimension and texture are extracted from each image region. These mathematical descriptors are automatically associated with intermediate semantic descriptors. The intermediate descriptors are used also for image categorization and for qualitative definition of semantic keywords in the user queries. For improving the initial query results, we apply a relevance feedback mechanism that uses the low -level descriptors of the images selected as relevant by user for producing the final query results. A support vector machine classifier can be learned from training data of relevance images and irrelevance images marked by users. Using the classifier and the semantic indexing, we implement a software system that can retrieve more images relevant to the query in the database efficiently.

Topic Map for Medical E-Learning

The paper presents original ways of using a modern concept - topic map - in medical e-learning. The topic map is mainly used for visualizing a thesaurus containing medical terms. The topic map is built and populated in an original manner, mapping an xml file that can be downloaded free, to an xtm file that contains the structure of the topic map. Only a part of the MeSH thesaurus was used, namely the part that includes the medical diagnosis’s names. The student can navigate through topic map depending on its interest subject, having in this way big advantages. The paper presents also how to use the topic map for semantic querying of a multimedia database with medical information and images. For retrieving the interest information this access path can be combined with another modern solution: the content-based visual query on the multimedia medical database. Combining these possibilities to access a database with medical data and images, allows students to see images and associated information in a simple and direct manner. The students are stimulated to learn, by comparing similar cases or by comparing cases that are visually similar, but with different diagnoses.

Database Kernel for Image Retrieval

This article presents a software tool that implements a dedicated multimedia database management server for managing alphanumerical and multimedia data collections from medical domain. An element of originality for this database management system (DBMS) is that along with classical operations for databases, it includes a series of algorithms used for extracting visual information from images (texture and color characteristics). The color histogram with 166 colors in HSV space represents the image color information. A vector with 12 values represents the texture information obtained by applying Gabor filters. The extracted data are stored in the database in a special data type called IMAGE, with a specific structure that can be used for visual queries. To increase the image retrieval speed, there are used some clustering algorithms.

Mapping Image Low-Level Descriptors to Semantic Concepts

Our goal is to organize the image contents semantically. In this paper, we propose a method to classify the images semantically, using the C-fuzzy algorithm to segment the natural scenes into perceptually uniform regions. The low-level characteristics that are taken into account are: color, texture, shape, absolute spatial arrangement, spatial coherency, and dimension. Since humans are the ultimate users of most image retrieval systems, it is important to organize the contents semantically, according to meaningful categories. This requires an understanding of the important semantic categories that humans use for image classification, and the extraction of meaningful image features that can discriminate between these categories. A lot of experiments, in which the human subjects had to group images into semantic categories and to explain the criteria for their choice, were realized. From these experiments, we identify the semantic categories (landscapes, animals, flowers, etc), the semantic indicators or intermediate descriptors and their visual characteristics.

Content-Based Image Query on Color Feature in the Image Databases Obtained from DICOM Files

The paper presents a qualitative study of the content-based image query process over the image databases resulting from the standard DICOM files produced by the medical devices used in the diagnosis process. There were implemented and studied several methods of transformation and quantization of the color space (HSV quantized at 166 colors, RGB quantized at 64 colors and CIE-LUV quantized at 512 colors) and also three methods for dissimilitude computing between the query and the target image (Euclidian distance, histograms intersection and quadratic distance between histograms). The algorithms that permit the extraction of the alphanumeric and imagistic information from the standard DICOM files and the algorithms for transforming and quantization of the color spaces are also presented. The experimental results displayed in a tabular and graphical form, permit the formulation of some useful conclusion, because the color medical images are more complex than the usual images

Improving an image retrieval system by integrating semantic features

To develop image navigation systems, we need tools to realize the semantic relationship between user and database. In this paper, it is presented an indexing schema of images and a simple semantic vocabulary that permits to the user to introduce the cognitive dimension in the retrieval process. A lot of experiments, in which the human subjects had to group images into semantic categories and to explain the criteria for their choice, were realized. From these experiments, we identify the semantic categories (landscapes, animals, flowers, etc), the semantic indicators or intermediate descriptors and their visual characteristics.

Color image segmentation applied to medical domain

The article presents two practical ways of using the automated color image segmentation in the medical field: for content-based region query and for tracking the time evolution of the disease in patients following a certain treatment. A known technique was used for automated color medical image segmentation - the color set back-projection algorithm. Our previous work in extraction of color regions from a database of nature images using the same algorithm showed promising results. The images are transformed from RGB to HSV color space, quantized at 166 colors and processed by the color set backprojection algorithm that allows the color region detection. The algorithm is studied from two points of view: complexity and the retrieval quality. The experiments that were made on a database with color endoscopy images from digestive tract have shown satisfying results for both applications that are important in practical medical use and medical teaching.

Rule-Based Methods for the Computer Assisted Diagnosis of Medical Images

In this paper we study a method for automatic diagnosis based on edoscopies of medical images. We develop algorithms that automatically generate rules that identify medical diagnosis. A semantic rule is a combination of semantic indicator values that identifies the image diagnosis. These rules are represented in Prolog and can be shared and modified depending on the updates in a respective domain. The experiments are realized on colour endoscopies from digestive apparatus, recording promising results.

Using the color set back-projection algorithm in retrieval and evaluation of endoscopic images from patients with peptic ulcer

This paper presents an original method of implementation of the color set back-projection algorithm, algorithm that allows the automated detection of the color regions from a color medical image. For the implementation of the color set back-projection algorithm, the image is transformed from RGB color space to HSV color space and quantized at 166 colors. At the end of this process, the color set of the image is obtained and used in color region detection. The software was tested on a gastroenterologic imagistic database that included 202 patients with peptic ulcer disease. One endoscopic image was taken at inclusion and another two at visit 1 and 2 for each patient, in order to evaluate the cicatrisation process under treatment. It was compared the percentage of correct diagnosis attended by using the classic observational method with the results obtained with the computer-based one in terms of reliability and reproducibility of this method. There were evaluated the advantages of using a computer-based retrieval system in terms of reducing the time spent on this operation. The inter-observer disagreement between human observer and the computer software was significantly low and the speed of the computerized method was higher, proving benefits in terms of time and cost efficiency.