Image bit enhancement method based on multilayer features of series neural network

A neural network and feature map technology, applied in the field of deep neural network, can solve the problems of fuzzy false contours, false contours that cannot be completely eliminated, low image visual quality, etc., and achieve stable image gradients, reduce network calculations, and high visual quality. Effect

Active Publication Date: 2020-02-14
TIANJIN UNIV
View PDF2 Cites 9 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The image bit enhancement algorithm based on a simple convolutional neural network proves that the deep learning network can learn more features, can effectively blur false contours and reconstruct high-bit images, but in large color transition areas, false contours cannot be completely eliminated , resulting in lower image visual quality

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 bit enhancement method based on multilayer features of series neural network
  • Image bit enhancement method based on multilayer features of series neural network
  • Image bit enhancement method based on multilayer features of series neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0034] The embodiment of the present invention proposes a multi-feature fusion convolutional neural network based on a variational autoencoder for image bit enhancement, and optimizes the network model through a gradient descent perceptual loss function. The method includes the following steps:

[0035] 101: Sintel for high bit lossless picture quality [9] 、UST-HK [10] , KODAK [11] The image in the database is preprocessed. First, the high-bit image is quantized to the low-bit image, and then the high-bit image and the low-bit image are calculated by pixel to obtain the residual image.

[0036] Among them, the Sintel database comes from a lossless animation short film, and the UST-HK and KODAK databases are real photos. Randomly select 1000 pictures in the Sintel database as the training set, all pictures in the UST-HK and KODAK databases and 50 pictures in the Sintel database except the training set as the test set.

[0037] 102: The present invention uses the improved VAE...

Embodiment 2

[0042] The scheme in embodiment 1 is further introduced below, see the following description for details:

[0043] 201: Since the Sintel database composed of animated images is completely generated by computer software, the images are free from noise interference, so the images in the Sintel database tend to have smoother color gradient structures, and the edges and textures in the images are also clearer. This near-ideal structural feature can help the neural network learn the features of smooth regions and edge structures, and help the model reconstruct the color gradient structure in the image and keep the outline relatively sharp. Therefore, the deep neural network proposed in this paper is trained with Sintel animation images . The UST-HK, KODAK database and part of Sintel composed of real photographs are used as test sets to verify the effect of the present invention.

[0044] In view of the image structure characteristics of the low bit depth image and the corresponding ...

Embodiment 3

[0069] Below in conjunction with concrete experimental data, the scheme in embodiment 1 and 2 is carried out effect assessment, see the following description for details:

[0070] 301: Data composition

[0071] The training set consists of 1000 images randomly selected from the Sintel database.

[0072] The test set consists of 50 images randomly selected by Sintel in addition to the training set and all images in the UST-HK and KODAK databases.

[0073] 302: Evaluation Criteria

[0074]The present invention mainly adopts two kinds of evaluation indicators to evaluate the reconstructed high-bit image quality:

[0075] PSNR (Peak Signal to Noise Ratio, Peak Signal to Noise Ratio) is the most widely used objective criterion for evaluating image quality. It is the mean square error between the original image and the comparison image relative to (2 n -1) 2 The logarithmic value of (that is, the square of the maximum value of the signal, where n is the number of bits). The la...

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 bit enhancement method based on multilayer features of a series neural network. The image bit enhancement method comprises the following steps: constructing a trainingset, quantizing a high-bit image of the training set into a low-bit image, solving a difference between the high-bit image and the low-bit image according to pixels to obtain a residual image, and performing zero filling on the low-bit image to obtain a zero-filled high-bit image; removing random variables in the VAE network, directly inputting a feature map generated by an encoder into a decoder, and establishing a deep learning network model on the basis of the feature map; adding a plurality of series jump connections into the network model, and transmitting each layer of feature map to all subsequent layers; inputting the zero-filling high-bit image into a deep learning network model to generate a residual image, and training a network by using an Adam optimizer; and quantizing the high-bit image of the test set into a low-bit image, inputting the zero-filling high-bit image into the network loaded with the training model parameters to generate a residual image, and adding the residual image and the low-bit image according to pixels to obtain a reconstructed high-bit image.

Description

technical field [0001] The invention relates to the field of deep neural networks, in particular to an image bit enhancement method based on multi-layer features of serial neural networks. Background technique [0002] With the development and penetration of the visual information industry, people's requirements for the visual quality provided by the display are also increasing. High-definition display and HDR (High Dynamic Range, High Dynamic Range) display can greatly expand the displayed brightness range, display more brightness and dark details, bring richer colors and more vivid and natural details to the picture, making the picture closer to what the human eye sees. Therefore, high-definition displays and HDR displays are gradually becoming mainstream devices in the market. [0003] However, limited by the current shooting equipment, each color channel of each pixel in most images and videos is stored with 8 bits, so each color channel can display up to 256 colors. ...

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): G06T5/00
CPCG06T5/007G06T2207/20081G06T2207/20084Y02D10/00
Inventor 于洁潇张春萍刘婧
Owner TIANJIN UNIV
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