IMIOL: A SYSTEM FOR INDEXING IMAGES BY THEIR SEMANTIC CONTENT BASED ON POSSIBILISTIC FUZZY CLUSTERING AND ADAPTIVE RESONANCE THEORY NEURAL NETWORKS LEARNING

摘要

Image databases are becoming large and of potential use in many areas, including medical diagnosis, astronomy, and the Web. These images, if analyzed, can reveal useful and potential information. Image indexing is the process of extracting and modeling the content of the image, the image data relationships, or other patterns not explicitly stored. Done this way, images could be indexed with the extracted knowledge, and thereby searching in these large databases for a particular information or pattern becomes more efficient and more reliable. For example, a medical doctor can search a medical database for already-diagnosed patient images having the same symptoms as the one at hand. Here we present a model for image indexing that bridges the gap between the visual content (or low-level descriptors) and the semantic content (or concepts). In our proposal, an image is modeled as being a set of objects, and each object is modeled with visual and semantics contents. Determination of the distinct objects (or segmentation) of the image is achieved through Possibilistic Fuzzy clustering. Thereafter, the visual content (namely, color, texture, and shape) of each object is computed using image processing techniques. Subsequently, each object is presented to the Concept Object Knowledge Base (COKB) for extraction of the semantic content using shape-based recognition. This knowledge base is constructed via neural learning with Adaptive Resonance Theory networks. Experimentation on standard large image databases reveal a good performance of our model.