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

Intrinsic image decomposition method based on jump layer frequency division and multi-scale identification of Unet

An intrinsic image and multi-scale technology, applied in the field of image processing, can solve the problem of high frequency components

Active Publication Date: 2020-08-21
NORTHWESTERN POLYTECHNICAL UNIV
View PDF13 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

On the other hand, adding frequency decomposition and frequency compression to the skip connection of the light map can not only get a more suitable feature map but also solve the problem of high frequency components in the light map

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
  • Intrinsic image decomposition method based on jump layer frequency division and multi-scale identification of Unet
  • Intrinsic image decomposition method based on jump layer frequency division and multi-scale identification of Unet
  • Intrinsic image decomposition method based on jump layer frequency division and multi-scale identification of Unet

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0126] (1) Build a training image sample library

[0127] Using the MPI image dataset, there are 9 categories of scenes commonly used in the MPI dataset, each category has two subcategories, and each category has 50 pictures. There are two segmentation methods when constructing the training image sample library, one is based on the image-split method, and the other is the scene-split method.

[0128] In the image-split method, in the 18 small categories of the image data set, half of the images are extracted for each small category, and each image size is 1024x436, and 10 small images with a size of 256x256 are randomly sampled in the image, and then the The small images are flipped horizontally so that each image gets 20 small images. The total number of training data sets is 9000 (18x25x20) small images with a size of 256x256, and the test data set uses the original image size, with a total of 450 (18x25) large images with a size of 1024x436.

[0129] In the Scene-split me...

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 intrinsic image decomposition method based on jump layer frequency division and multi-scale identification of Unet. According to the method, a Unet-based generative adversarial network is constructed, the network is composed of a generator and a discriminator, the generator is used for decomposing an image into a reflection image and an illumination image, and the discriminator is used for discriminating whether the image is true or false and guiding the generator to generate a faked image. The network designed by the invention effectively alleviates the problem caused by the fact that the encoder features are directly sent to the decoder. On one hand, the constraint of frequency decomposition is added to the jump connection of the Unet of the reflection graph, sothat the network learns the importance degree of different features, and more suitable features are obtained; and on the other hand, frequency decomposition and frequency compression are added in jump connection of the illumination image, so that a more appropriate feature image can be obtained, and the problem of multiple high-frequency components in the illumination image is solved.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to an intrinsic image decomposition method. Background technique [0002] Image recognition applications have appeared in various places of life, such as face recognition, target tracking, driverless driving, etc. However, in the imaging process, due to the influence of many environmental factors such as light intensity, light incident angle, and shadow occlusion, the imaging effect may be poor, which makes image recognition difficult and the accuracy is reduced. One way to solve this problem is to extract features that do not change with environmental factors in multi-modal images, that is, intrinsic images. Intrinsic image refers to the inherent characteristics of an object independent of environmental factors, including color, texture, material, etc. These inherent characteristics will not change with changes in environmental factors. By separating the intrinsic info...

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): G06N3/04G06N3/08G06T5/00
CPCG06N3/08G06T2207/20081G06T2207/20084G06N3/048G06N3/045G06T5/70
Inventor 蒋晓悦方阳王鼎李煜祥冯晓毅
Owner NORTHWESTERN POLYTECHNICAL UNIV