A Probabilistic Analysis of Sparse Coded Feature Pooling and Its Application for Image Retrieval

Abstract

Feature coding and pooling as a key component of image retrieval have been widely studied over the past several years. Recently sparse coding with max-pooling is regarded as the state-of-the-art for image classification. However there is no comprehensive study concerning the application of sparse coding for image retrieval. In this paper, we first analyze the effects of different sampling strategies for image retrieval, then we discuss feature pooling strategies on image retrieval performance with a probabilistic explanation in the context of sparse coding framework, and propose a modified sum pooling procedure which can improve the retrieval accuracy significantly. Further we apply sparse coding method to aggregate multiple types of features for large-scale image retrieval. Extensive experiments on commonly-used evaluation datasets demonstrate that our final compact image representation improves the retrieval accuracy significantly.

A location-aware embedding technique for accurate landmark recognition

“…2 shows some examples of the newly-added query images. According to [22], to increase the recognition accuracy, we resized the database images to 320x240 (or 240x320 if they are rotated) during the creation of VLAD vectors for the database images.…”

“…However, as shown in Table 2, the locVLAD approach can obtain superior results on top1 even using a smaller vocabulary (4281 wrt 10M for [7]). Conversely, in the case of sparse coding in [22], a vocabulary composed of 8k 64 − D SURF descriptors plus 1k color descriptors is used. In order to perform a fair comparison, also locVLAD is tested with a vocabulary of the same size, i.e 64 * 8k + 36 * 1k = 548k.…”

“…Zhang et al [22] implemented a method based on sparse coding and by using max pooling in alternative to sum pooling (used in the traditional VLAD implementation).…”

Feature Encodings and Poolings for Action and Event Recognition: A Comprehensive Survey

“…For 3D encodings, three popular approaches were surveyed, i.e., spatiotemporal grids encoding [62,63], STLPC encoding [64], and spatiotemporal VLAD encoding [65]. For general poolings, three popular approaches were surveyed, i.e., sum pooling [66][67][68][69][70], average pooling [71][72][73][74][75][76], and max pooling [77][78][79][80]. For particular poolings, another three popular approaches were surveyed, i.e., stochastic pooling [81], semantic pooling [82], and multi-scale pooling [83][84][85][86].…”

“…For example, Peng et al [66] analyzed action recognition performance among several bag of visual words and fusion methods, where they adopted sum pooling and power l 2 -normalization for pooling and normalization strategy. Zhang et al [67] gave a probabilistic interpretation why the max pooling was usually better than sum pooling in the context of sparse coding framework for image retrieval applications, since max pooling tended to increase the discrimination of the similarity measurement than sum pooling. Besides, they proposed a modified sum pooling method, improving the retrieval accuracy significantly over the max pooling strategy.…”