[Paper Review] Image classification based on support vector machine and the fusion of complementary features

Huilin Gao, Wenjie Chen+1 more|arXiv (Cornell University)|Nov 5, 2015

Advanced Image and Video Retrieval Techniques22 references4 citations

TL;DR

This paper proposes a novel image classification method that fuses three complementary features—PHOW, PHOC, and PHOG—using an adaptive feature-weighted SVM to improve accuracy. By enhancing the traditional BOW model with dense sampling and an improved K-means clustering for visual dictionary construction, the approach achieves 7%–17% higher classification accuracy than conventional BOW methods on the Caltech 101 dataset.

ABSTRACT

Image Classification based on BOW (Bag-of-words) has broad application prospect in pattern recognition field but the shortcomings are existed because of single feature and low classification accuracy. To this end we combine three ingredients: (i) Three features with functions of mutual complementation are adopted to describe the images, including PHOW (Pyramid Histogram of Words), PHOC (Pyramid Histogram of Color) and PHOG (Pyramid Histogram of Orientated Gradients). (ii) The improvement of traditional BOW model is presented by using dense sample and an improved K-means clustering method for constructing the visual dictionary. (iii) An adaptive feature-weight adjusted image categorization algorithm based on the SVM and the fusion of multiple features is adopted. Experiments carried out on Caltech 101 database confirm the validity of the proposed approach. From the experimental results can be seen that the classification accuracy rate of the proposed method is improved by 7%-17% higher than that of the traditional BOW methods. This algorithm makes full use of global, local and spatial information and has significant improvements to the classification accuracy.

Motivation & Objective

• To address the limitations of traditional Bag-of-Words (BOW) methods in image classification, which suffer from single-feature representation and low accuracy.
• To enhance feature representation by combining global, local, and spatial information through complementary features: PHOW, PHOC, and PHOG.
• To improve the BOW model by introducing dense sampling and an enhanced K-means clustering algorithm for visual dictionary construction.
• To develop an adaptive feature-weighted SVM framework that optimizes the fusion of multiple complementary features for improved categorization performance.

Proposed method

• The method employs three complementary features: PHOW (Pyramid Histogram of Words) for local texture and spatial layout, PHOC (Pyramid Histogram of Color) for color distribution, and PHOG (Pyramid Histogram of Oriented Gradients) for edge and gradient patterns.
• A dense sampling strategy is applied to extract more robust visual features from images, increasing feature diversity and coverage.
• An improved K-means clustering algorithm is used to build a more discriminative visual vocabulary, enhancing the quality of the BOW representation.
• An adaptive feature-weight adjustment mechanism is integrated with SVM to dynamically assign optimal weights to each feature type based on their contribution to classification.
• The fused feature representation is fed into an SVM classifier, which learns to distinguish between image categories using the combined information from all three feature types.
• The entire pipeline is trained and evaluated on the Caltech 101 benchmark dataset to validate performance gains.

Experimental results

Research questions

• RQ1 Can the fusion of multiple complementary features—PHOW, PHOC, and PHOG—improve image classification accuracy beyond single-feature BOW methods?
• RQ2 How does dense sampling and an improved K-means clustering method enhance the visual vocabulary construction in the BOW framework?
• RQ3 To what extent can adaptive feature weighting in an SVM framework improve classification performance when combining diverse visual features?
• RQ4 Does the integration of global, local, and spatial features lead to a significant accuracy gain on standard image classification benchmarks?

Key findings

• The proposed method achieves a classification accuracy improvement of 7% to 17% over traditional BOW-based methods on the Caltech 101 dataset.
• The fusion of PHOW, PHOC, and PHOG features captures complementary information, significantly enhancing the discriminative power of the visual representation.
• The use of dense sampling and an improved K-means clustering method results in a more robust and discriminative visual vocabulary compared to standard sparse sampling.
• The adaptive feature-weighted SVM approach effectively balances the contribution of each feature type, leading to more stable and accurate classification outcomes.
• The method demonstrates strong generalization by leveraging global (PHOC), local (PHOW), and spatial gradient (PHOG) information simultaneously.
• The experimental results confirm that the proposed approach outperforms conventional BOW methods, validating the effectiveness of feature fusion and improved feature representation.