Hierarchical Matching Pursuit for Image Classification: Architecture and Fast Algorithms
نویسندگان
چکیده
Extracting good representations from images is essential for many computer vision tasks. In this paper, we propose hierarchical matching pursuit (HMP), which builds a feature hierarchy layer-by-layer using an efficient matching pursuit encoder. It includes three modules: batch (tree) orthogonal matching pursuit, spatial pyramid max pooling, and contrast normalization. We investigate the architecture of HMP, and show that all three components are critical for good performance. To speed up the orthogonal matching pursuit, we propose a batch tree orthogonal matching pursuit that is particularly suitable to encode a large number of observations that share the same large dictionary. HMP is scalable and can efficiently handle full-size images. In addition, HMP enables linear support vector machines (SVMs) to match the performance of nonlinear SVMs while being scalable to large datasets. We compare HMP with many state-of-the-art algorithms including convolutional deep belief networks, SIFT based single layer sparse coding, and kernel based feature learning. HMP consistently yields superior accuracy on three types of visual recognition problems: object recognition (Caltech-101), scene recognition (MIT-Scene), and static event recognition (UIUC-Sports).
منابع مشابه
Performance Evaluation of Local Detectors in the Presence of Noise for Multi-Sensor Remote Sensing Image Matching
Automatic, efficient, accurate, and stable image matching is one of the most critical issues in remote sensing, photogrammetry, and machine vision. In recent decades, various algorithms have been proposed based on the feature-based framework, which concentrates on detecting and describing local features. Understanding the characteristics of different matching algorithms in various applications ...
متن کاملSpectral-spatial classification of hyperspectral images by combining hierarchical and marker-based Minimum Spanning Forest algorithms
Many researches have demonstrated that the spatial information can play an important role in the classification of hyperspectral imagery. This study proposes a modified spectral–spatial classification approach for improving the spectral–spatial classification of hyperspectral images. In the proposed method ten spatial/texture features, using mean, standard deviation, contrast, homogeneity, corr...
متن کاملKernel Descriptors in comparison with Hierarchical Matching Pursuit
In this paper a state of the art sparse coding algorithm for image classification, namely Hierachical Matching Pursuit(HMP), is compared to state of the art algorithms using kernel methods(Efficient Match Kernels, Kernel Descriptors and Hierarchical Kernel Descriptors). HMP is faster and achieves slightly better results than the other algorithms when run over several test-sets. But on the downs...
متن کاملMultiresolution segmentation-based image coding with hierarchical data structures
This paper presents two multiresolution segmentation-based algorithms for low bit rate image compression using hierarchical data structures. The segmentation is achieved with quadtree and BSP-tree hierarchical data structures and the encoding is performed by using the projection pursuit (matching pursuit) with a finite dictionary of spline functions with various degrees of smoothness. Compariso...
متن کاملFast Least Square Matching
Least square matching (LSM) is one of the most accurate image matching methods in photogrammetry and remote sensing. The main disadvantage of the LSM is its high computational complexity due to large size of observation equations. To address this problem, in this paper a novel method, called fast least square matching (FLSM) is being presented. The main idea of the proposed FLSM is decreasing t...
متن کامل