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

Image quality classification method based on dual-channel deep parallel convolutional network

A convolutional network and image quality technology, applied in the field of image processing, can solve problems such as limited effect of image quality features, failure to fully consider the impact of image semantic information on image quality, and low classification accuracy, so as to improve classification performance and reduce complexity. Sexuality, the effect of eliminating influence

Inactive Publication Date: 2018-12-07
XIDIAN UNIV
View PDF6 Cites 27 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The disadvantage of the first type of method is that it requires a lot of experience and relevant background knowledge. Due to the highly abstract nature of human image quality evaluation behavior, the image quality features designed with reference to artificial experience have limited effects, and they also have high requirements at the engineering level.
The second type of method automatically extracts features from images. Although it solves the defects of artificially designed features, the existing deep learning models do not fully consider the impact of image semantic information on image quality, resulting in low classification accuracy.

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 quality classification method based on dual-channel deep parallel convolutional network
  • Image quality classification method based on dual-channel deep parallel convolutional network
  • Image quality classification method based on dual-channel deep parallel convolutional network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0028] Since manual design of features requires a lot of experience and relevant background knowledge, it is difficult to manually design image quality features, and the designed features are relatively one-sided, and the classification effect is limited. However, the existing deep learning models do not fully consider the impact of image semantic information on image quality. The classification accuracy is not high. The present invention analyzes and studies these problems, and proposes an image quality classification method based on a dual-channel deep parallel convolution network, see figure 1 , including the following steps:

[0029] (1) Select image samples: Select all images in the database related to image quality evaluation as experimental data, that is, image samples. There are M types of images with different contents in the database, and each type of image contains high-quality image samples and low-quality image samples.

[0030] The image sample data used in the...

Embodiment 2

[0043] The image quality classification method based on dual-channel depth parallel convolution network is the same as embodiment 1, and the construction of dual-channel depth parallel convolution network described in step (3) includes the following steps:

[0044] (3a) Build a seven-layer single-channel deep convolutional network: the single-channel deep convolutional network does not have a parallel convolutional structure, see figure 2 , the specific structure is:

[0045] The first four layers are sequentially connected convolutional layers, the last three layers are sequentially connected fully connected layers, and the fourth convolutional layer is connected to the fifth fully connected layer. Specifically: the first layer is a convolution layer that includes convolution processing, pooling processing, and local response normalization processing; the second layer is a convolution layer that includes convolution processing, pooling processing, and local response normaliz...

Embodiment 3

[0056] The image quality classification method based on dual-channel deep parallel convolutional network is the same as embodiment 1-2, and the two data preprocessing methods described in step (4) and step (6) refer to:

[0057] The first preprocessing method first adjusts the size of the training data set or the image to be tested, scales the original size of the training data set or the image to be tested to a 256*256 size image, and then randomly crops from the scaled image An image block with a size of 227*227 is output as the input image of the first channel of the dual-channel deep parallel convolutional network.

[0058] In the second preprocessing method, an image block with a size of 227*227 is randomly cropped directly from the original size training data set or the image to be tested, and used as the input image of the second channel of the dual-channel deep parallel convolutional network.

[0059] The purpose of randomly cropping image blocks in the present inventi...

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 an image quality classification method based on a dual-channel deep parallel convolutional network. The image quality classification method solves the problems of difficulty inartificial construction of quality features and loss of image detailed information in convolutional neural network input normalization. The image quality classification method comprises the implementation steps of: selecting image samples; constructing a training image data set; establishing the dual-channel deep parallel convolutional network; preprocessing the training data set; training the dual-channel deep parallel convolutional network; performing data preprocessing on an image to be tested; and giving out an image quality classification result. The image quality classification method designs the new dual-channel deep parallel convolutional network, inputs a global image and local images via two channels separately, comprehensively considers the global information and detailed information of the images, designs a parallel convolution structure for the global image, can eliminate the influence of image semantic information on image quality classification, realizes comprehensive,reasonable and detailed image quality classification, and is widely used in computer vision and image aesthetic quality evaluation.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to image evaluation and classification, in particular to an image quality classification method based on a dual-channel deep parallel convolutional network. The present invention can be used to evaluate the quality of natural images, and can be further applied to distinguish the aesthetic quality of images. Background technique [0002] At present, with the development and popularization of digital devices such as mobile phones and cameras, photography has gradually attracted more and more attention and interest. Due to the popularity of these devices, the number of images has also increased at an explosive rate. Image quality classification is an important research issue in image processing and image analysis. This technology attempts to use various attribute features to deeply mine image quality-related characteristics, realize accurate classification of image qual...

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
IPC IPC(8): G06T7/00G06N3/04
CPCG06T7/0002G06T2207/30168G06N3/045
Inventor 高新波李恒达路文余颖何立火
Owner XIDIAN UNIV
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