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.


Histogram of sum pooling results.(a) Visual codes C3, C4 generated by distinctive patches on the background. (b) Visual codes C4 generated by frequent patches on the background.
© Copyright Policy
Related In: Results  -  Collection

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

pone.0131721.g001: Histogram of sum pooling results.(a) Visual codes C3, C4 generated by distinctive patches on the background. (b) Visual codes C4 generated by frequent patches on the background.

Mentions: Dense sampling extracts the patches uniformly which may contain lots of repeated and redundant information on the clean background. The repeated patches can be divided into two categories: 1) distinctive patches are denoted as those repeated patches which are present in a little part of train images and 2) frequent patches are denoted as those repeated patches which are present in most of the training images. Fig 1 illustrates the feature coding step of dataset images. Dense patches of training images are extracted and a codebook with five visual codes is trained. The sum pooling results of images on the codebook generate a histogram. The repeated patches on the left images fall into various bins. The visual code C3 and C4 have a strong discrimination. However, the repeated patches on the right images fall into the same bin. The code C4 has low discrimination. It is obviously to be seen that the distinctive patches on the background can contribute to improving the retrieval performance while frequent patches will not. It is similar to some extent with thought of IDF in BOW model, which is not included in sparse coding framework.


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)

Histogram of sum pooling results.(a) Visual codes C3, C4 generated by distinctive patches on the background. (b) Visual codes C4 generated by frequent patches on the background.
© Copyright Policy
Related In: Results  -  Collection

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

pone.0131721.g001: Histogram of sum pooling results.(a) Visual codes C3, C4 generated by distinctive patches on the background. (b) Visual codes C4 generated by frequent patches on the background.
Mentions: Dense sampling extracts the patches uniformly which may contain lots of repeated and redundant information on the clean background. The repeated patches can be divided into two categories: 1) distinctive patches are denoted as those repeated patches which are present in a little part of train images and 2) frequent patches are denoted as those repeated patches which are present in most of the training images. Fig 1 illustrates the feature coding step of dataset images. Dense patches of training images are extracted and a codebook with five visual codes is trained. The sum pooling results of images on the codebook generate a histogram. The repeated patches on the left images fall into various bins. The visual code C3 and C4 have a strong discrimination. However, the repeated patches on the right images fall into the same bin. The code C4 has low discrimination. It is obviously to be seen that the distinctive patches on the background can contribute to improving the retrieval performance while frequent patches will not. It is similar to some extent with thought of IDF in BOW model, which is not included in sparse coding framework.

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.