PicSOM: self-organizing maps for content-based image retrieval

Abstract

Digital image libraries are becoming more common and widely used as more visual information is produced at a rapidly growing rate. Content-based image retrieval is an important approach to the problem of processing this increasing amount of data It is based on automatically extracted features from the content of the images, such as color, texture, shape, and structure. We have started a project to study methods for content-based image remkval using the Self-organizing Map (SOM) as the image similarity scoring method Our image rem*evd system, named Pic-SOM. can be seen as a SOM-based approach to relevance feedback which is a form of supervised learning to adjust the subsequent queries based on the user's responses during the information rem-eval session. In PicSOM, a separate Tree Structured SOM (TS-SOM) is trained for each feature vector type in use. The system then adapts to the user's preferences by returning her more images from those SOMs where her responses have been most densely mapped.

BALAS: Empirical Bayesian learning in the relevance feedback for image retrieval

“…The intuition was to emphasize more on the features that best cluster the positive samples and maximize the separation between the positive and negative samples. To achieve this goal, Kohonen's Learning Vector Quantization (LVQ) algorithm [42] and the tree-structured self-organizing map (TS-SOM) [15] were used for dynamic data clustering during the relevance feedback.…”

“…For example, Laaksonen et al [15] applied TS-SOMs to index images along different feature dimensions such as color and texture. Positive and negative examples were mapped to positive and negative impulses and a low-pass filtering was applied to generate a map to implicitly reveal relative importance of different features, as a "good" map always keeps positive examples well clustered while negative examples are scattered away.…”

Comparing discriminating transformations and SVM for learning during multimedia retrieval

“…Some use Kohonen's Learning Vector Quantization (LVQ) algorithm [33] or Self-organizing Map (SOM) [13] for dynamic data clustering during relevance feedback. Laaksonen et al [13] uses TS(Tree-Structured)-SOMs to index the images along different features.…”

“…Laaksonen et al [13] uses TS(Tree-Structured)-SOMs to index the images along different features. Positive and negative examples are mapped to positive and negative impulses on the maps and a low-pass operation on the maps is argued to implicitly reveal the relative importance of different features because a "good" map will keep positive examples cluster while negative examples scatter away.…”

Interactive Content-Based Retrieval Using Pre-computed Object-Object Similarities

“…Other recent research is inspired by machine learning methods. Self-organising maps [11] and support vector machines [4,15] are employed to solve the problems of CBIR. Many existing retrieval systems rely on active participation of the searcher in the retrieval process, which is known as relevance feedback [13].…”