Abstract:The codebook model has been developed as an effective means for image classification. However, the inherent operation of assigning visual words to image feature vectors in traditional codebook approaches causes serious ambiguities in image classification. In particular, the nearest word may not be the best fit to a feature, and multiple words may be equally appropriate for one specific feature. To resolve these ambiguities, we propose a novel visual codebook model based on the n-dimensional fuzzy geometry (n-D FG) theory, where all visual words and features are modeled as fuzzy points in the n-D FG space, and appropriate uncertainty is introduced to each fuzzy point to enhance the representation capacity. This n-D FG-codebook model not only inherits advantages from the fuzzy set theory, but also facilitates the analysis and determination of the relationship between visual words and features in geometric form. By explicitly taking into account the ambiguities, we propose a novel measure of similarity between the visual words and fuzzy features. Following the proposed codebook model and the novel similarity measure, we develop two useful image classification algorithms by modifying popular image coding algorithms (i.e. SPM and LLC). Finally, experimental results demonstrate that the classification accuracy of the proposed algorithms is dramatically improved for a standard large-scale image database. For example, with a codebook size of 256, the proposed algorithms achieve similar performance as traditional algorithms with a codebook size of 1024, indicating that the proposed algorithms reduce the computational cost by 75% while achieving almost identical classification accuracy to traditional algorithms. Thus, the proposed algorithms represent a more efficient and appropriate scheme for big image data. (C) 2015 Elsevier Ltd. All rights reserved.

Image Classification by Codebook Updating via Joint i-Pat Topic Model Feedback Wai

Image Classification Method by Combining Multi-features and Sparse Coding

Enhanced Random Forest with Image/Patch-Level Learning for Image Understanding

Shrinking Encoding with Two-Level Codebook Learning for Fine-Grained Fish Recognition

Randomness and Sparsity Induced Codebook Learning with Application to Cancer Image Classification

Sparse Codebook Model of Local Structures for Retrieval of Focal Liver Lesions Using Multiphase Medical Images

Codebook Reconstruction with Word Correlation Feedback Mechanism.

Online Codebook Reweighting Using Pairwise Constraints for Image Classification

A novel visual codebook model based on fuzzy geometry for large-scale image classification

Image classification using adapted codebook

Discriminative Spatial Codebook Generation for Image Classification

Discriminative Codebook Learning for Web Image Search.

Improvements in image categorization using codebook ensembles

Effective Image Representation Based on Bi-Layer Visual Codebook

Unifying Discriminative Visual Codebook Generation with Classifier Training for Object Category Recognition

Multi-layer orthogonal visual codebook for image classification

Contextual Aerial Image Categorization Using Codebook.

Codebook Reconstruction with Holistic Information Fusion

A Unified Supervised Codebook Learning Framework for Classification.

Visual word coding based on difference maximization.

An Image Classification Method Based on Optimized Fuzzy Bag-of-words Model