Abstract:Recently, relevance feedback (RF) in content-based image retrieval (CBIR) has been implemented as an online binary classifier to separate the positive samples from the negative samples, where both sets of samples are labeled by the user. In many applications, it is reasonable to assume that all the positive samples are alike and thus that the region of the feature space occupied by the positive samples can be described by a single hypersurface. However, for the negative samples, previous RF methods either treat each one of the negative samples as an isolated point or assume the whole negative set can be described by a single convex hypersurface. In this paper, we argue that these treatments of the negative samples are not sound. Our belief is all positive samples are included in a set and the negative samples split into a small number of subsets, each one of which has a simple distribution. Therefore, we first cluster the negative samples into several groups; for each such negative group, we build a marginal convex machine (MCM) subclassifier between it and the single positive group which results in a series of subclassifiers. These subclassifiers are then incorporated into a biased MCM (BMCM) for RF. Experiments were carried out to prove the advantages of BMCM-based RF over previous methods for RF.

Nonparametric discriminant analysis in relevance feedback for content-based image retrieval

Direct Kernel Biased Discriminant Analysis: A New Content-Based Image Retrieval Relevance Feedback Algorithm.

Kernel full-space biased discriminant analysis

A Direct Method to Solve the Biased Discriminant Analysis in Kernel Feature Space for Content Based Image Retrieval

Learning an Image Manifold for Retrieval

Image retrieval based on incremental subspace learning

Biased Discriminant Euclidean Embedding for Content-Based Image Retrieval.

A Survey on Relevance Feedback Techniques in Content-based Medical Image Retrieval

Asymmetric Bagging and Random Subspace for Support Vector Machines-Based Relevance Feedback in Image Retrieval

Designing a new deep convolutional neural network for content-based image retrieval with relevance feedback

Adaptive quasiconformal kernel discriminant analysis

Negative Samples Analysis in Relevance Feedback

Which Components Are Important for Interactive Image Searching?

Similarity Measure Learning for Image Retrieval Using Binary Component Discriminating Function

Relevance Feedback in Content-Based Image Retrieval: Bayesian Framework, Feature Subspaces, and Progressive Learning

Multi-Class Discriminant Learning for Image Retrieval.

Kernel Generalized Nonlinear Discriminant Analysis Algorithm For Pattern Recognition

Data-Dependent Kernel Discriminant Analysis for Feature Extraction and Classification

Using High Dimensional Indexes to Support Relevance Feedback Based Interactive Images Retrieval

Feature Subspaces Extraction for Content-Based Image Retrieval

Evaluation of relevance feedback methods for 3D model retrieval