Image super-resolution method and system thereof

A super-resolution and low-resolution technology, applied in the field of image processing, can solve problems such as practical application limitations, increased computational complexity, and large training redundancy, and achieve the effect of reducing information redundancy, preventing gradient explosion, and enriching image information.

Inactive Publication Date: 2018-05-01
北京花开影视制作有限公司 +2
View PDF3 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] However, the existing neural network-based image super-resolution technology has large training redundancy and is only applicable to a single-scale problem, resulting in a sharp increase in the amount of calculations, causing computational waste, and limiting its practical application.

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 super-resolution method and system thereof
  • Image super-resolution method and system thereof
  • Image super-resolution method and system thereof

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0062] figure 1 The main flowchart of the method is shown, including the following steps:

[0063] Step S110, preprocessing the input image.

[0064] Preprocessing involves upscaling the input image.

[0065] Including the following sub-steps:

[0066] Step S1101, input a low-resolution picture;

[0067] Step S1102, read the size of the high-resolution picture to be obtained;

[0068] Step S1103: Enlarge the low-resolution input picture based on quadratic cubic interpolation, so that the low-resolution input picture has the same size as the output picture.

[0069] Step S120, using the convolution network to realize the extraction and representation of the pixel block of the input picture.

[0070] Pre-construct a convolutional network of N (N is an integer greater than 1) layer, and use the convolutional network to realize the extraction and representation of the pixel block of the input image, including the following sub-steps:

[0071] Step S1201, using the first i (i...

Embodiment 2

[0080] The above describes how to use a convolutional network to convert a low-resolution image into a high-resolution image. In the current deep network training, since it is necessary to use the surrounding pixels to infer the central pixel, each time a convolutional layer of the convolutional network is added , the size of the feature map is reduced. For example: the size of the input image is (n+1)x(n+1), and when the receptive field of the network is nxn, the output image will be 1x1, where the receptive field refers to the output characteristics of each layer of the convolutional neural network The size of the area mapped by the pixels on the feature map on the original image.

[0081] This processing method prevents the pixels on the border of the image from using the surrounding pixels. The current processing method is to cut the border pixels. This is obviously not suitable when the axial area of ​​​​the image is very large, because at this time, after cutting the su...

Embodiment 3

[0085] The image super-resolution method is described above through Embodiment 1 and Embodiment 2, and the low-resolution picture is converted into a high-resolution picture. The following describes the training method of the convolutional network used in the above-mentioned embodiment 1 in conjunction with the accompanying drawings, such as figure 2 shown, including the following steps:

[0086] Step S210, building a sample training database;

[0087] The sample training library includes multiple sets of training samples, and each set of training samples includes a low-resolution picture x and a high-resolution picture y corresponding to the low-resolution picture x.

[0088] Step S220, read a set of low-resolution pictures and their corresponding high-resolution pictures from the sample training database.

[0089] Step S230, calculating the loss function according to the low-resolution picture and the high-resolution picture; including the following sub-steps:

[0090] St...

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 picture super-resolution method. The method comprises the following steps of: preprocessing an input picture; extracting and expressing a pixel block of the input picture byusing a convolutional network; obtaining an image detail difference value of the input picture by using the convolutional network; and realizing picture reconstruction by using a convolutional layer.According to the picture super-resolution method, the convolutional network is trained to only learn picture details, so that the information redundancy is decreased; a gradient is controlled in a certain range by adoption of gradient shearing, so that gradient explosion is prevented; and furthermore, a boundary filling operation is carried out on the periphery of the input picture, so that edge pixels of the picture also can correctly learn the picture details, and pooling layers which are commonly used in super-resolution calculation can be removed, thereby ensuring the consistency between network input and output dimensionalities.

Description

technical field [0001] The present application relates to the field of image processing, in particular to an image super-resolution method and system thereof. Background technique [0002] Super-resolution technology refers to converting low-resolution (Low Resolution, LR) images into high-resolution (High Resolution, HR) images through certain algorithms. Because high-resolution images have higher pixel density, more detailed information, and more delicate picture quality, they are generally welcomed by people. [0003] The super-resolution technology of a single image is to use a low-resolution image and use a conversion algorithm to convert it into a high-resolution image. This kind of conversion is widely used in the field of computer vision, such as the field of security monitoring, medical field, image transmission field, etc., when it is necessary to see more detailed information in the image, this conversion technology will be used. [0004] Existing transformation...

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): G06T3/40G06T5/50
CPCG06T3/4053G06T3/4007G06T5/50G06T2207/20081G06T2207/20084G06T2207/20221G06T2207/20228
Inventor 赵明明李小波
Owner 北京花开影视制作有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap