An efficient key point quantization algorithm for large scale image retrieval

Abstract

We focus on the problem of large-scale near duplicate image retrieval. Recent studies have shown that local image features, often referred to as key points, are effective for near duplicate image retrieval. The most popular approach for key point based image matching is the clustering-based bagof-words model. It maps each key point to a visual word in a code-book that is constructed by a clustering algorithm, and represents each image by a histogram of visual words. Despite its success, there are two main shortcomings of the clustering-based bag-of-words model: (i) it is computationally expensive to cluster millions of key points into thousands of visual words; (ii) there is no theoretical analysis on the performance of the bag-of-words model. We propose a new scheme for key point quantization that addresses these shortcomings. Instead of clustering, the proposed scheme quantizes each key point into a binary vector using a collection of randomly generated hyper-spheres, and a bag-ofwords model is constructed based on such randomized quantization. Our theoretical analysis shows that the resulting image similarity provides an upper bound for the similarity based on the optimal partial matching between two sets of key points. Empirical study on a database of 100, 000 images shows that the proposed scheme is not only more efficient but also more effective than the clustering-based approach for near duplicate image retrieval.

Scalable Feature Matching by Dual Cascaded Scalar Quantization for Image Retrieval

“…Comparison is conducted with seven recent retrieval algorithms on local feature quantization, including visual vocabulary tree [4], Hamming embedding [6], soft assignment [5], binary SIFT [21], product quantization with inverted file structure [16], vector of locally aggregated descriptor [22], and random seeding approach [23]. The experiments demonstrate that our method achieves competitive performance in terms of accuracy, efficiency, and memory usage.…”

Robust local and global shape context for tattoo image matching

“…SIFT was chosen to find a "visually similar" image, hence improving the image retrieval accuracy with respect to just using low level attributes. SIFT was also combined with a Bag-Of-Words (BOW) model in [6], [7], and [8]. In [6] the computational complexity of SIFT feature clustering was examined when the features are quantized in the BOW model.…”

“…SIFT was also combined with a Bag-Of-Words (BOW) model in [6], [7], and [8]. In [6] the computational complexity of SIFT feature clustering was examined when the features are quantized in the BOW model. Instead of using clustering, a random seed algorithm was proposed to make the quantization process faster and more accurate.…”

An Augmented Reality Setup with an Omnidirectional Camera Based on Multiple Object Detection

“…Since these descriptors are high dimensional vectors such as 128 dimensions with larger computational costs, Sivic and Zisserman [7] quantized the SIFT descriptors by a single depth based k-means tree to achieve run-time object retrieval throughout a movie database. These quantized descriptors are usually called visual words, which are generated by a vocabulary tree [2], [8], [9], [10]. Visual words are collected to depict an object and exploited to a classification of objects by ignoring a position of each visual word [11].…”