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

Zhang Y, Chen J, Huang X, Wang Y - PLoS ONE (2015)

Bottom Line: 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.

View Article: PubMed Central - PubMed

Affiliation: School of computer science & technology, Beijing Institute of Technology, Beijing, China.

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.

No MeSH data available.


Distance between similar and dissimilar images on UKB dataset.(a) The probability density of L1 norm distance with sum pooling. (b) The probability density of L1 norm distance with max pooling.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4489107&req=5

pone.0131721.g002: Distance between similar and dissimilar images on UKB dataset.(a) The probability density of L1 norm distance with sum pooling. (b) The probability density of L1 norm distance with max pooling.

Mentions: As we can see from the L1 norm of expectation, max pooling tends to increase the discrimination of the similarity measurement than sum pooling, especially with the increasing of pooling cardinality N. Therefore similar and dissimilar images can be more easily separated with max pooling than sum pooling with the growth of pooling cardinality N. In order to proof this, we experimentally calculate the L1 norm distance of similar and dissimilar images. Fig 2 shows the statistical frequency of the L1 norm distance with max and sum pooling schemes. The solid histogram stands for the probability density of L1 norm distance between similar images, while dashed histogram stands for dissimilar images. As shown in the statistical histogram in Fig 2, max pooling can easily separate similar images from dissimilar images with the increasing of pooling cardinality N. On the other hand, we can easily get that the retrieval performance of sum pooling and max pooling will both benefit from the growth of k codewords.


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

Zhang Y, Chen J, Huang X, Wang Y - PLoS ONE (2015)

Distance between similar and dissimilar images on UKB dataset.(a) The probability density of L1 norm distance with sum pooling. (b) The probability density of L1 norm distance with max pooling.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4489107&req=5

pone.0131721.g002: Distance between similar and dissimilar images on UKB dataset.(a) The probability density of L1 norm distance with sum pooling. (b) The probability density of L1 norm distance with max pooling.
Mentions: As we can see from the L1 norm of expectation, max pooling tends to increase the discrimination of the similarity measurement than sum pooling, especially with the increasing of pooling cardinality N. Therefore similar and dissimilar images can be more easily separated with max pooling than sum pooling with the growth of pooling cardinality N. In order to proof this, we experimentally calculate the L1 norm distance of similar and dissimilar images. Fig 2 shows the statistical frequency of the L1 norm distance with max and sum pooling schemes. The solid histogram stands for the probability density of L1 norm distance between similar images, while dashed histogram stands for dissimilar images. As shown in the statistical histogram in Fig 2, max pooling can easily separate similar images from dissimilar images with the increasing of pooling cardinality N. On the other hand, we can easily get that the retrieval performance of sum pooling and max pooling will both benefit from the growth of k codewords.

Bottom Line: 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.

View Article: PubMed Central - PubMed

Affiliation: School of computer science & technology, Beijing Institute of Technology, Beijing, China.

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.

No MeSH data available.