Unlock instant, AI-driven research and patent intelligence for your innovation.

Multi-scale feature fusion image hash retrieval method and system based on deep learning

A multi-scale feature, fusion image technology, applied in still image data retrieval, still image data indexing, still image data clustering/classification, etc., can solve problems such as increased difficulty, avoid information loss, ensure consistency, The effect of increasing the cross-entropy loss

Inactive Publication Date: 2019-11-08
INSPUR ARTIFICIAL INTELLIGENCE RES INST CO LTD SHANDONG CHINA
View PDF3 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Due to the presentation of massive high-dimensional data, the difficulty of quickly and accurately retrieving the images that users want from specific databases also increases.

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
  • Multi-scale feature fusion image hash retrieval method and system based on deep learning
  • Multi-scale feature fusion image hash retrieval method and system based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0036] Such as figure 1 As shown, the multi-scale feature fusion image hash retrieval method based on deep learning of the present invention specifically includes the following steps:

[0037] S1. Training the deep network: Input the training data into the built deep network, and obtain the corresponding hash code and hash function through the backpropagation algorithm.

[0038] Training the deep network includes the convolutional network part in the front section, the multi-scale feature fusion part and the loss function part after the convolutional layer.

[0039] In the convolutional network part, in order to learn image features, the input image is convolved, pooled and activated, and input to the last convolutional layer.

specific example

[0040] The multi-scale feature fusion part after the convolutional layer solves the inconsistency of the input image size, copies the output of the convolutional layer of multiple channels, and divides the area, and performs the maximum pooling operation on the divided area. After pooling, the The data corresponding to the same division scale are added together to obtain the feature vector, and the obtained feature vectors are spliced ​​together and input to the fully connected layer for learning of the hash layer. Among them, a deep convolutional neural network structure with multi-scale feature fusion can be constructed. Specifically, a ResNet network with strong learning ability can be used. The specific example is as follows: for the output of the convolutional layer of d channels, we copy it into n copies, and for each copy, we divide it into regions. The region division can be set according to specific needs, for example, the first One area is divided into one area, the ...

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 discloses a multi-scale feature fusion image hash retrieval method and a system based on deep learning, and belongs to the technical field of image retrieval in computer vision. The invention discloses a multi-scale feature fusion image hash retrieval method based on deep learning. The method comprises the following steps of S1, training a deep network: inputting training data into the built deep network, and obtaining a corresponding hash code and a hash function through a back propagation algorithm; S2, inputting the image data in the image database into the trained deep network, and obtaining and storing a hash code database; S3, during image retrieval, inputting a specific image into the deep network trained in the step S1 to obtain a corresponding retrieval image hash code; and S4, performing exclusive-OR operation on the hash code database and the retrieval image hash code. According to the multi-scale feature fusion image hash retrieval method based on deep learning, the dimension consistency of data input into a full connection layer can be guaranteed, and the method has good application and popularization value.

Description

technical field [0001] The invention relates to the technical field of image retrieval in computer vision, and specifically provides a multi-scale feature fusion image hash retrieval method and system based on deep learning. Background technique [0002] With the rapid development of Internet technology, the number of pictures on the network has increased dramatically. How to retrieve the images that users want with high accuracy from massive high-dimensional image data is a problem that has been widely concerned and studied by researchers. Due to the presentation of massive high-dimensional data, the difficulty of quickly and accurately retrieving the images that users want from a specific database also increases. Hash technology has been widely used in the field of image retrieval due to its advantages of high retrieval speed and small storage consumption. The key to the hash retrieval method is to map the high-dimensional original data features into a low-dimensional bin...

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): G06F16/51G06F16/55G06K9/62
CPCG06F16/51G06F16/55G06F18/24G06F18/253
Inventor 张雨柔李锐于治楼
Owner INSPUR ARTIFICIAL INTELLIGENCE RES INST CO LTD SHANDONG CHINA