Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Image classification method based on locality constraints and saliency

A classification method and local technology, applied in the field of computer vision, to achieve the effect of improved classification results and strong discriminative power

Inactive Publication Date: 2018-04-27
BEIJING UNION UNIVERSITY
View PDF4 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The problem with the hard-assigned encoding method is that it is very sensitive to distortion errors in the dictionary; the hard-assigned method only selects the nearest visual word, ignoring other related visual words
This method is only suitable for pictures of multi-target image scenes, and is not suitable for all pictures for classification

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Image classification method based on locality constraints and saliency
  • Image classification method based on locality constraints and saliency
  • Image classification method based on locality constraints and saliency

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0050] In this embodiment, the image classification method based on locality constraints and saliency includes two parts: a training process 100 and a testing process 120 .

[0051] In the training process 100, step 101 is executed to select a certain number of images to construct a training set D, and the corresponding class label set is L. Step 102 is executed to extract the local feature descriptors of each image in the training set to form a set Y of local feature descriptors. Step 103 is executed to cluster the set Y of local feature descriptors in the training set to obtain a visual word dictionary B, where M is the number of visual words. Step 104 is executed to extract local feature descriptors for each training image, and the set is X. Execute step 105, perform locality-constrained encoding calculations on each training image, and the encoding formula based on locality constraints is: where x i Indicates the i-th local feature descriptor; b j Indicates the jth vi...

Embodiment 2

[0055] When encoding an image, it is intuitively understood that the significant part of the encoded value should be given a greater weight to highlight the role of this part of the feature in the image representation. The description of the idea is as follows: figure 2 Shown is the original image, Figure 2a Shown is the saliency map corresponding to the original image. exist Figure 2b The black dots in represent the local feature descriptors in the background to be encoded, such as Figure 2c As shown, the diagram shows the coding situation of black dots in background pixels before the introduction of salience, such as Figure 2d As shown, the figure shows the encoding of background pixel black dots after the introduction of salience. It can be seen that for the background pixel black dots, due to the low saliency value, the encoding value after the introduction of salience is higher than that before the introduction of salience. Coded values ​​have been reduced. exis...

Embodiment 3

[0057]The experimental image databases include Oxford 17 flower database, Oxford 102 flower database, Caltech 101, Caltech 256 and UIUC8 database. The local features used in the experiment are 128-dimensional SIFT feature descriptors, the sampling interval is 8 pixels, and the area blocks around the descriptors are 16×16 pixels. Each image library obtains a dictionary containing 400 visual words through the k_means method; 30 images are randomly selected for each type of image as training, and the remaining images are used as testing. In the experiment, the DRFI salient region extraction method is used to obtain the saliency map of the image. In the experiment, the parameter β is equal to 10, and the number of nearest neighbors is 5. The experimental results are shown in Table 1. In the table, the LLC method is used to encode and calculate the Oxford 17 flower library, Oxford 102 flower library, Caltech 101, Caltech 256 and UIUC8 library, and the obtained visual word coding ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides an image classification method based on locality constraints and saliency, which comprises the step of performing training on an image. The image classification method further comprises the steps of extracting local feature factors of each test image to form a set X; performing locality-constrained coding calculation each the test image; performing saliency calculation on each test image; fusing saliency values and coded values of pixels to obtain saliency based coded values; performing a feature pooling operation on each test image by adopting a spatial pyramid model; and inputting a feature f into an SVM classifier for testing to obtain a classification category label. The method considers locality constraints of a near neighbor and takes the distance between the near neighbor and the local feature as a basis for coding on the one hand, and considers the role of the saliency in image coding on the other hand.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an image classification method based on locality constraints and saliency. Background technique [0002] Image classification methods generally go through the feature encoding process of images. The original BoF method uses a hard-assigned method to encode local descriptors. Hard assignment refers to the assignment of local features to the nearest visual word in the visual dictionary, the assigned visual word is coded as 1, and the remaining visual words are coded as 0. The problem with the hard-assigned coding method is that it is very sensitive to distortion errors in the dictionary; the hard-assigned method only selects the nearest visual word, ignoring other related visual words. For the shortcomings of the hard assignment method, soft quantization is to describe a feature descriptor with multiple visual words. The advantage of soft quantization is that the concept...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/62
CPCG06F18/23G06F18/2411G06F18/25
Inventor 梁晔马楠李华丽宋恒达陈强昝艺璇蒋元胡路明
Owner BEIJING UNION UNIVERSITY
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com