WO2017075768A1 - 一种基于字典匹配的图像超分辨率重建方法及装置 - Google Patents
一种基于字典匹配的图像超分辨率重建方法及装置 Download PDFInfo
- Publication number
- WO2017075768A1 WO2017075768A1 PCT/CN2015/093765 CN2015093765W WO2017075768A1 WO 2017075768 A1 WO2017075768 A1 WO 2017075768A1 CN 2015093765 W CN2015093765 W CN 2015093765W WO 2017075768 A1 WO2017075768 A1 WO 2017075768A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- filter
- layer
- image block
- image
- resolution image
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 230000003321 amplification Effects 0.000 claims abstract description 6
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 6
- 238000001914 filtration Methods 0.000 claims description 53
- 238000012549 training Methods 0.000 claims description 36
- 238000013507 mapping Methods 0.000 claims description 27
- 238000012545 processing Methods 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 238000003064 k means clustering Methods 0.000 claims description 5
- 230000004044 response Effects 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/772—Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Definitions
- the present application relates to a method and apparatus for image super-resolution reconstruction based on dictionary matching.
- Super-Resolution also known as upsampling and image magnification, refers to the recovery of high-resolution, sharp images from low-resolution images.
- Super-resolution is one of the fundamental problems in the field of image and video processing. It has a very wide application prospect in the fields of medical image processing, image recognition, digital photo processing, and high-definition television.
- kernel-based interpolation algorithms such as bilinear interpolation, spline interpolation, and so on.
- the interpolation algorithm generates continuous data by using known discrete data, so blurring, aliasing, and the like occur, and the image restoration effect is not good.
- the present application provides a method and apparatus for image super-resolution reconstruction based on dictionary matching, which can improve the quality of reconstructed high-resolution images.
- the present application provides a method for image super-resolution reconstruction based on dictionary matching, comprising: establishing a matching dictionary library; inputting an image block to be reconstructed into a multi-layer line filter network, and extracting the to-be-reconstructed a local feature of the image block to be reconstructed in the image; searching for a local feature of the low-resolution image block having the highest similarity with the local feature of the image to be reconstructed from the matching dictionary library; searching for the matching dictionary library
- the local feature of the low-resolution image block with the highest similarity is located in the residual value of the joint sample; the local feature of the low-resolution image block with the highest similarity is interpolated and amplified, and the residual value is added to obtain the reconstruction After the high resolution image block.
- the present application provides an image super-resolution reconstruction apparatus based on dictionary matching, comprising: an establishing unit for establishing a matching dictionary library; and an extracting unit for inputting the image block to be reconstructed into multiple layers.
- a matching unit is configured to search for a low resolution image with the highest similarity to the local feature of the image to be reconstructed from the matching dictionary library a local feature of the block; a search unit for finding a residual value of the joint sample in which the local feature of the low-resolution image block having the highest similarity is located in the matching dictionary library; and a difference amplifying unit configured to compare the similarity
- the local feature of the highest-resolution low-resolution image block is subjected to interpolation and amplification; the reconstruction unit is configured to add the local feature of the low-resolution image block obtained by the difference amplification unit to the residual found by the searching unit Value, get the reconstructed high resolution image block.
- the method and device for image super-resolution reconstruction based on dictionary matching establishes a matching dictionary library, inputs the image to be reconstructed into a multi-layer line filter network, extracts local features of the image to be reconstructed, and searches for a matching dictionary library. Describe the local features of the low-resolution image block with the highest local similarity of the reconstructed image, and find the residual value of the joint sample of the local feature with the highest similarity in the matching dictionary database.
- the local features of the low-resolution image block with the highest similarity are interpolated and amplified, and the residual values are added to obtain a reconstructed high-resolution image block.
- the local features of the image to be reconstructed extracted by the multi-layer line filter network have higher precision. Therefore, when matching with the matching dictionary database, the matching degree is higher, and the reconstructed image quality is also better. Thus, embodiments of the present application can greatly improve the quality of reconstructed high resolution images.
- FIG. 1 is a flowchart of a method for image super-resolution reconstruction based on dictionary matching according to the present application
- FIG. 2 is a schematic flow chart of step 101 in FIG. 1;
- FIG. 3 is a schematic diagram of a local feature extraction process of an image of a multi-layer line filter network in an embodiment of the present application
- FIG. 4 is a schematic diagram of a filtering process of the present application.
- FIG. 5 is a schematic structural diagram of an image super-resolution reconstruction apparatus based on dictionary matching according to the present application.
- Figure 6 is a schematic structural view of the establishing unit in Figure 5;
- Figure 7 is a schematic view showing the structure of the extraction unit in Figure 5 .
- an image super-resolution reconstruction method and apparatus based on dictionary matching is provided, which can improve the quality of the reconstructed high-resolution image.
- Embodiment 1 is a diagrammatic representation of Embodiment 1:
- FIG. 1 is a flowchart of a method according to Embodiment 1 of the present application.
- a method for image super-resolution reconstruction based on dictionary matching may include the following steps:
- the step 102 may include: Step 1:
- the multi-layer line filter network includes a filtering layer, and the first stage filter of the filtering layer filters the input image block to be reconstructed by using N different size line filtering windows to obtain a corresponding N filtered images are output to a next stage filter, the filtered image comprising: line features of the image, where N is an integer greater than one.
- Step 2 The second stage filter of the filter layer filters the N filtered images output by the first stage filter by using M different linear filter windows to obtain corresponding M ⁇ N filtered images, where M Is an integer greater than 1.
- Step 3 repeatedly output all the filtered images obtained by each level filter to the next-stage filter, and the next-stage filter uses a plurality of line filtering windows to filter all the filtered images output by the upper-level filter respectively until the last one After filtering the stage filter, all the filtered images obtained are output to the mapping layer of the multi-layer line filter network.
- Step 4 The mapping layer performs binarization processing on all the filtered images of the filtering layer, and outputs the same to the output layer of the multi-layer line filter network.
- Step 5 The output layer performs concatenation and output on the binarized filtered image output by the mapping layer, that is, obtains the local feature of the input image block to be reconstructed.
- the local features of the image block to be reconstructed extracted by the multi-layer line filter network have higher precision. Therefore, when matching with the matching dictionary database, the matching degree is higher, and the reconstructed image quality is also good.
- the extraction of image features by the multi-layer line filter network proposed in the embodiments of the present application is also applicable to reconstruction methods based on manifold learning or sparse representation.
- step 101 of this embodiment specifically includes the following steps:
- 500,000 joint samples can be randomly selected from a known training image library, and 1024 joint samples are clustered from the above 500,000 clusters using a K-means clustering algorithm, and the 1024 joint samples are used as The atom of the dictionary constitutes a matching dictionary.
- the local feature extraction process of the image by the multi-layer line filter network includes:
- the multi-layer line filter network includes a filter layer, and the first stage filter of the filter layer utilizes N A different size of the line filtering window filters the image in the input network to obtain corresponding N filtered images, and outputs them to the next stage filter.
- the filtered image includes: a line feature of the image, and N is an integer greater than one.
- the second stage filter of the filter layer filters the N filtered images output by the first stage filter by using M different linear filter windows to obtain corresponding M ⁇ N filtered images.
- M is an integer greater than one.
- step S2 all the filtered images obtained by each level of the filter are repeatedly output to the next-stage filter, and the next-stage filter uses a plurality of line filtering windows to respectively perform all the filtered images output by the upper-stage filter.
- the filtering process is performed until the last stage filter is filtered, and all the obtained filtered images are output to the mapping layer of the multilayer line filter network.
- the number of repetitions of filtering can be performed by setting a number of stages of filters in advance according to actual needs.
- mapping layer binarizes all the filtered images of the filter layer and outputs them to the output layer of the multi-layer line filter network.
- the output layer connects and outputs the binarized filtered image outputted by the mapping layer to obtain a local feature of the image;
- the input image of the filter network is a whole image, and the output layer first performs a block histogram on each of the binarized processed filtered images output by the mapping layer, and then performs concatenation and output, thereby obtaining local features of the image.
- the first step the construction of a multi-layer line filter network.
- a line filter group with different directions of one bandwidth is used to extract line features in an image, and the line filter response is calculated as follows:
- the coordinates of the pixel points on the line L k are specifically defined as follows:
- (i 0 , j 0 ) is the central pixel point coordinate of the local block and S k is the slope of the line L k .
- the filter bank calculates the pixel values in the different directions on the line in the filter window, and then selects the pixel value in the direction with the smallest value and acts as a filter response.
- Figure 4 depicts the network structure of a multi-layer line filter network.
- P in the input network can be the entire image.
- the multi-layer line filter network is used to extract global statistical features. It can be a partial image block for extracting local features.
- the network structure mainly includes a filtering layer, a mapping layer, and an output layer.
- the filter layer contains multiple levels of line filtering operations. This embodiment is described by taking a two-stage line filtering operation as an example.
- the first stage of filtering filters the input P using a total of N1 different sizes of the above line filters LF:
- the results obtained by filtering all the windows in the first-stage filter are output to the next-stage filter, that is, the second-stage filter.
- the output is filtered with the first stage of p i 1 .
- the second stage of filtering uses each output of the first stage of filtering as an input, filtering using a total of N2 differently sized line filter windows:
- the output is filtered with p ij 2 second stage. It can be seen from Fig. 4 that by repeating the multi-stage filtering operation, more filtered images are obtained, and the filter network can be extended to a higher layer.
- the filter layer After the filter layer is the mapping layer, the filter layer will finally filter and output multiple image features to the mapping layer, and the mapping layer binarizes each output of the filtering layer, as shown in formula (5), and then multiple binarized The output is combined into a map:
- LB is a local binarization operation, defined as follows:
- x is the pixel point in the middle of the current filtering window
- x p is the adjacent pixel point of the pixel
- the entire image is divided into a plurality of partial image blocks, and the filtered images of the respective partial image blocks are respectively obtained, and thus, in the output layer, First, a block histogram is made for each output of the mapping layer, and then connected as a global statistical feature of the image P.
- mapping layer output is joined as an image local feature.
- the present application also provides an image super-resolution reconstruction method based on dictionary matching, establishes a matching dictionary library, inputs the image to be reconstructed into a multi-layer line filter network, extracts local features of the image to be reconstructed, and searches for and from the matching dictionary library. Describe the local features of the low-resolution image block with the highest local similarity of the reconstructed image, and find the residual value of the joint sample of the local feature with the highest similarity in the matching dictionary database.
- the local features of the low-resolution image block with the highest similarity are interpolated and amplified, and the residual values are added to obtain a reconstructed high-resolution image block.
- the local features of the image to be reconstructed extracted by the multi-layer line filter network have higher precision. Therefore, when matching with the matching dictionary database, the matching degree is higher, and the reconstructed image quality is also better. Thus, embodiments of the present application can greatly improve the quality of reconstructed high resolution images.
- Embodiment 2 is a diagrammatic representation of Embodiment 1:
- an embodiment of the present application provides an image super-resolution reconstruction apparatus based on dictionary matching, including: an establishing unit 30, configured to create a matching dictionary library, and further includes:
- the extracting unit 31 is configured to input the image block to be reconstructed into the multi-layer line filter network, and extract local features of the image block to be reconstructed.
- the matching unit 32 is configured to search, from the matching dictionary library, local features of the low-resolution image block with the highest similarity to the local features of the image block to be reconstructed.
- the searching unit 33 searches for a residual value of the joint sample in which the local feature of the highest-resolution low-resolution image block is located in the matching dictionary library.
- the difference amplifying unit 34 is configured to perform interpolation and amplification on local features of the low resolution image block with the highest similarity.
- the reconstruction unit 35 is configured to add the local feature of the low-resolution image block that is amplified by the difference amplifying unit 34 to the residual value found by the searching unit to obtain the reconstructed high-resolution image block.
- the establishing unit 30 specifically includes:
- the acquiring module 30A is configured to collect a plurality of high-resolution image blocks, and respectively downsample the plurality of high-resolution image blocks to obtain a low-resolution image block corresponding to each of the high-resolution image blocks, one
- the high resolution image block and the low resolution image block corresponding to the high resolution image block form a pair of training samples.
- the subtraction module 30B is configured to subtract the image of the high resolution image block in each pair of training samples and the image of the low resolution image block by interpolation to obtain a residual value of the training sample.
- the extraction module 30C is configured to input a low resolution image block of each pair of training samples into the multi-layer line filter network, and extract local features of the low resolution image blocks of each pair of training samples.
- the splicing module 30D is configured to splicing the local features of the low-resolution image blocks of each pair of training samples and the residual values of the training samples as joint samples of the training samples.
- the training module 30E is configured to train a plurality of joint samples by using K-means clustering to obtain a matching dictionary library.
- the extracting unit 31 specifically includes:
- the first stage filter 31A is configured to filter the input image block to be reconstructed by using N different size line filter windows to obtain corresponding N filtered images, and output to the next stage filter, where the filtered image includes : a line feature of the image, wherein N is an integer greater than one;
- the second stage filter 31B is configured to separately filter the N filtered images output by the first stage filter by using M different size line filter windows to obtain corresponding M ⁇ N filtered images, where M is greater than An integer of 1;
- the filtering module 31C is configured to repeatedly output all the filtered images obtained by each level of the filter to the next-stage filter, and the next-stage filter uses a plurality of line filtering windows to filter all the filtered images output by the upper-level filter respectively. Until the last stage filter is filtered, all the obtained filtered images are output to the mapping layer of the multi-layer line filter network;
- mapping layer 31D configured to perform binarization processing on all filtered images of the filtering layer, and output to an output layer of the multi-layer line filter network
- the output layer 31E is configured to filter the filtered image after the binarization output of the mapping layer
- the lines are connected and output, that is, the local features of the input image block to be reconstructed are obtained.
- Each preset size filtering window linearly filters pixel points (i 0 , j 0 ) located in the middle of the filtering window in the image by using a plurality of linear filters in different directions in the window region, and the response formula as follows:
- the coordinates of the pixel, the line L k is defined as follows:
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims (10)
- 一种基于字典匹配的图像超分辨率重建方法,其特征在于,包括:建立匹配字典库;将待重建图像块输入到多层线滤波器网络,提取所述待重建图像块的局部特征;从所述匹配字典库中寻找与所述待重建图像块的局部特征相似度最高的低分辨率图像块的局部特征;寻找在所述匹配字典库中,所述相似度最高的低分辨率图像块的局部特征所在联合样本的残差值;对所述相似度最高的低分辨率图像块的局部特征进行插值放大,加上所述残差值,获得重建后的高分辨率图像块。
- 如权利要求1所述的基于字典匹配的图像超分辨率重建方法,其特征在于,所述建立匹配字典库包括:采集多个高分辨率图像块,分别对所述多个高分辨率图像块进行降采样,得到与每个所述高分辨率图像块对应的低分辨率图像块,一个高分辨率图像块以及与所述高分辨率图像块对应的低分辨率图像块组成一对训练样本;将每对训练样本中的所述高分辨率图像块与所述低分辨率图像块进行插值放大后的图像相减,得到所述训练样本的残差值;将每对训练样本的低分辨率图像块输入多层线滤波器网络,提取每对训练样本的低分辨率图像块的局部特征;将所述每对训练样本的低分辨率图像块的局部特征以及所述训练样本的残差值拼接起来作为所述训练样本的联合样本;使用K均值聚类对多个联合样本进行训练,得到匹配字典库。
- 如权利要求1或2所述的基于字典匹配的图像超分辨率重建方法,其特征在于,所述多层线滤波器网络对图像的局部特征提取过程包括:步骤一:多层线滤波器网络包括滤波层,所述滤波层的第一级滤波器利用N个不同大小的线滤波窗口对输入网络中的图像进行滤波,得到对应的N个滤波图像,并输出到下一级滤波器,所述滤波图像包括:所述图像的线特征,其中N为大于1的整数;步骤二:所述滤波层的第二级滤波器利用M个不同大小的线滤波窗口分别对第一级滤波器输出的所述N个滤波图像进行滤波,得到对应的 M×N个滤波图像,其中M为大于1的整数;步骤三:重复将每级滤波器得到的所有滤波图像输出至下一级滤波器,下一级滤波器利用多个线滤波窗口对上级滤波器输出的所有滤波图像分别进行滤波处理,直至最后一级滤波器滤波完毕,得到的所有滤波图像输出至多层线滤波器网络的映射层;步骤四:所述映射层对所述滤波层的所有滤波图像进行二值化处理,并输出至多层线滤波器网络的输出层;步骤五:若所述多层线滤波器网络的输入图像为局部图像块,所述输出层对所述映射层输出的二值化处理后的滤波图像进行衔接并输出,即得到所述图像的局部特征;若所述多层线滤波器网络的输入图像为整幅图像,所述输出层先对所述映射层输出的二值化处理后的各个滤波图像分别做分块直方图,再进行衔接并输出,即得到所述图像的局部特征。
- 如权利要求3所述的基于字典匹配的图像超分辨率重建方法,其特征在于,所述将待重建图像块输入到多层线滤波器网络,提取所述待重建图像块的局部特征包括:步骤一:多层线滤波器网络包括滤波层,所述滤波层的第一级滤波器利用N个不同大小的线滤波窗口对输入的待重建图像块进行滤波,得到对应的N个滤波图像,并输出到下一级滤波器,所述滤波图像包括:所述图像的线特征,其中N为大于1的整数;步骤二:所述滤波层的第二级滤波器利用M个不同大小的线滤波窗口分别对第一级滤波器输出的所述N个滤波图像进行滤波,得到对应的M×N个滤波图像,其中M为大于1的整数;步骤三:重复将每级滤波器得到的所有滤波图像输出至下一级滤波器,下一级滤波器利用多个线滤波窗口对上级滤波器输出的所有滤波图像分别进行滤波处理,直至最后一级滤波器滤波完毕,得到的所有滤波图像输出至多层线滤波器网络的映射层;步骤四:所述映射层对所述滤波层的所有滤波图像进行二值化处理,并输出至多层线滤波器网络的输出层;步骤五:所述输出层对所述映射层输出的二值化处理后的滤波图像进行衔接并输出,即得到所述输入的待重建图像块的局部特征。
- 一种基于字典匹配的图像超分辨率重建装置,其特征在于,包括:建立单元,用于建立匹配字典库;提取单元,用于将待重建图像块输入到多层线滤波器网络,提取所述重建图像块的局部特征;匹配单元,用于从所述匹配字典库中寻找与所述待重建图像块的局部特征相似度最高的低分辨率图像块的局部特征;寻找单元,寻找在所述匹配字典库中,所述相似度最高的低分辨率图像块的局部特征所在联合样本的残差值;差值放大单元,用于对所述相似度最高的低分辨率图像块的局部特征进行插值放大;重建单元,用于将所述差值放大单元进行放大后的低分辨率图像块的局部特征加上所述寻找单元寻找到的残差值,获得重建后的高分辨率图像块。
- 如权利要求7所述的图像超分辨率重建装置,其特征在于,所述建立单元具体包括:采集模块,用于采集多个高分辨率图像块,分别对所述多个高分辨率图像块进行降采样,得到与每个所述高分辨率图像块对应的低分辨率图像块,一个高分辨率图像块以及与所述高分辨率图像块对应的低分辨率图像块组成一对训练样本;减法模块,用于将每对训练样本中的所述高分辨率图像块与所述低分辨率图像块进行插值放大后的图像相减,得到所述训练样本的残差值;提取模块,用于将每对训练样本的低分辨率图像块输入多层线滤波器网络,提取每对训练样本的低分辨率图像块的局部特征;拼接模块,用于将所述每对训练样本的低分辨率图像块的局部特征以及所述训练样本的残差值拼接起来作为所述训练样本的联合样本;训练模块,用于使用K均值聚类对多个联合样本进行训练,得到匹配字典库。
- 如权利要求7或8所述的图像超分辨率重建装置,其特征在于,所述提取单元具体包括:第一级滤波器,用于利用N个不同大小的线滤波窗口对输入的待重建图像块进行滤波,得到对应的N个滤波图像,并输出到下一级滤波器,所述滤波图像包括:所述图像的线特征,其中N为大于1的整数;第二级滤波器,用于利用M个不同大小的线滤波窗口分别对第一级滤波器输出的所述N个滤波图像进行滤波,得到对应的M×N个滤波图像,其中M为大于1的整数;滤波模块,用于重复将每级滤波器得到的所有滤波图像输出至下一级滤波器,下一级滤波器利用多个线滤波窗口对上级滤波器输出的所有滤波图像分别进行滤波处理,直至最后一级滤波器滤波完毕,得到的所有滤波图像输出至多层线滤波器网络的映射层;映射层,用于对所述滤波层的所有滤波图像进行二值化处理,并输出至多层线滤波器网络的输出层;输出层,所述输出层用于对所述映射层输出的二值化处理后的滤波图像进行衔接并输出,即得到所述输入的待重建图像块的局部特征。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US15/749,554 US10339633B2 (en) | 2015-11-04 | 2015-11-04 | Method and device for super-resolution image reconstruction based on dictionary matching |
PCT/CN2015/093765 WO2017075768A1 (zh) | 2015-11-04 | 2015-11-04 | 一种基于字典匹配的图像超分辨率重建方法及装置 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2015/093765 WO2017075768A1 (zh) | 2015-11-04 | 2015-11-04 | 一种基于字典匹配的图像超分辨率重建方法及装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2017075768A1 true WO2017075768A1 (zh) | 2017-05-11 |
Family
ID=58661464
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2015/093765 WO2017075768A1 (zh) | 2015-11-04 | 2015-11-04 | 一种基于字典匹配的图像超分辨率重建方法及装置 |
Country Status (2)
Country | Link |
---|---|
US (1) | US10339633B2 (zh) |
WO (1) | WO2017075768A1 (zh) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596838A (zh) * | 2018-05-09 | 2018-09-28 | 北京环境特性研究所 | 一种基于学习的单帧人脸图像超分方法及装置 |
CN108710950A (zh) * | 2018-05-11 | 2018-10-26 | 上海市第六人民医院 | 一种图像量化分析方法 |
CN108776954A (zh) * | 2018-06-26 | 2018-11-09 | 北京字节跳动网络技术有限公司 | 用于生成图像的方法和装置 |
CN108986029A (zh) * | 2018-07-03 | 2018-12-11 | 南京览笛信息科技有限公司 | 文字图像超分辨率重建方法、系统、终端设备及存储介质 |
CN108989731A (zh) * | 2018-08-09 | 2018-12-11 | 复旦大学 | 一种提高视频空间分辨率的方法 |
CN109615576A (zh) * | 2018-06-28 | 2019-04-12 | 西安工程大学 | 基于级联回归基学习的单帧图像超分辨重建方法 |
CN109741254A (zh) * | 2018-12-12 | 2019-05-10 | 深圳先进技术研究院 | 字典训练及图像超分辨重建方法、系统、设备及存储介质 |
CN111709442A (zh) * | 2020-05-07 | 2020-09-25 | 北京工业大学 | 一种面向图像分类任务的多层字典学习方法 |
CN111951167A (zh) * | 2020-08-25 | 2020-11-17 | 深圳思谋信息科技有限公司 | 超分辨率图像重建方法、装置、计算机设备和存储介质 |
WO2021093620A1 (en) * | 2019-11-15 | 2021-05-20 | Huawei Technologies Co., Ltd. | Method and system for high-resolution image inpainting |
CN113158928A (zh) * | 2021-04-27 | 2021-07-23 | 浙江云奕科技有限公司 | 一种基于图像识别的混凝土试块防造假方法 |
WO2022087778A1 (zh) * | 2020-10-26 | 2022-05-05 | 深圳大学 | 一种基于多层耦合映射的低分辨率图像识别方法 |
CN114492450A (zh) * | 2021-12-22 | 2022-05-13 | 马上消费金融股份有限公司 | 文本匹配方法及装置 |
CN114972043A (zh) * | 2022-08-03 | 2022-08-30 | 江西财经大学 | 基于联合三边特征滤波的图像超分辨率重建方法与系统 |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10410398B2 (en) * | 2015-02-20 | 2019-09-10 | Qualcomm Incorporated | Systems and methods for reducing memory bandwidth using low quality tiles |
US10701394B1 (en) * | 2016-11-10 | 2020-06-30 | Twitter, Inc. | Real-time video super-resolution with spatio-temporal networks and motion compensation |
CN110147864B (zh) * | 2018-11-14 | 2022-02-22 | 腾讯科技(深圳)有限公司 | 编码图案的处理方法和装置、存储介质、电子装置 |
CN109766863B (zh) * | 2019-01-18 | 2022-09-06 | 南京邮电大学 | 一种基于局部和稀疏非局部正则的人脸图像超分辨率方法 |
CN110033410B (zh) * | 2019-03-28 | 2020-08-04 | 华中科技大学 | 图像重建模型训练方法、图像超分辨率重建方法及装置 |
CN111476714B (zh) * | 2020-03-30 | 2022-10-28 | 清华大学 | 基于psv神经网络的跨尺度图像拼接方法及装置 |
CN111696042B (zh) * | 2020-06-04 | 2023-06-27 | 四川轻化工大学 | 基于样本学习的图像超分辨重建方法 |
CN112598575B (zh) * | 2020-12-22 | 2022-05-03 | 电子科技大学 | 一种基于特征处理的图像信息融合及超分辨率重建方法 |
CN112669216B (zh) * | 2021-01-05 | 2022-04-22 | 华南理工大学 | 一种基于联邦学习的并行空洞新结构的超分辨率重构网络 |
CN112950470B (zh) * | 2021-02-26 | 2022-07-15 | 南开大学 | 基于时域特征融合的视频超分辨率重建方法及系统 |
CN114067110A (zh) * | 2021-07-13 | 2022-02-18 | 广东国地规划科技股份有限公司 | 一种实例分割网络模型的生成方法 |
CN114418853B (zh) * | 2022-01-21 | 2022-09-20 | 杭州碧游信息技术有限公司 | 基于相似图像检索的图像超分辨率优化方法、介质及设备 |
CN117011196B (zh) * | 2023-08-10 | 2024-04-19 | 哈尔滨工业大学 | 一种基于组合滤波优化的红外小目标检测方法及系统 |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102142137A (zh) * | 2011-03-10 | 2011-08-03 | 西安电子科技大学 | 基于高分辨率字典的稀疏表征图像超分辨重建方法 |
CN105389778A (zh) * | 2015-11-04 | 2016-03-09 | 北京大学深圳研究生院 | 一种基于字典匹配的图像超分辨率重建方法及装置 |
-
2015
- 2015-11-04 WO PCT/CN2015/093765 patent/WO2017075768A1/zh active Application Filing
- 2015-11-04 US US15/749,554 patent/US10339633B2/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102142137A (zh) * | 2011-03-10 | 2011-08-03 | 西安电子科技大学 | 基于高分辨率字典的稀疏表征图像超分辨重建方法 |
CN105389778A (zh) * | 2015-11-04 | 2016-03-09 | 北京大学深圳研究生院 | 一种基于字典匹配的图像超分辨率重建方法及装置 |
Non-Patent Citations (1)
Title |
---|
CHEN, HUAHUA ET AL.: "Image Super-resolution Reconstruction Based on Residual Error", JOURNAL OF IMAGE AND GRAPHICS, vol. 18, no. 1, 31 January 2013 (2013-01-31), pages 43 - 45 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596838B (zh) * | 2018-05-09 | 2022-02-18 | 北京环境特性研究所 | 一种基于学习的单帧人脸图像超分方法及装置 |
CN108596838A (zh) * | 2018-05-09 | 2018-09-28 | 北京环境特性研究所 | 一种基于学习的单帧人脸图像超分方法及装置 |
CN108710950A (zh) * | 2018-05-11 | 2018-10-26 | 上海市第六人民医院 | 一种图像量化分析方法 |
CN108776954A (zh) * | 2018-06-26 | 2018-11-09 | 北京字节跳动网络技术有限公司 | 用于生成图像的方法和装置 |
CN109615576A (zh) * | 2018-06-28 | 2019-04-12 | 西安工程大学 | 基于级联回归基学习的单帧图像超分辨重建方法 |
CN108986029B (zh) * | 2018-07-03 | 2023-09-08 | 南京览笛信息科技有限公司 | 文字图像超分辨率重建方法、系统、终端设备及存储介质 |
CN108986029A (zh) * | 2018-07-03 | 2018-12-11 | 南京览笛信息科技有限公司 | 文字图像超分辨率重建方法、系统、终端设备及存储介质 |
CN108989731B (zh) * | 2018-08-09 | 2020-07-07 | 复旦大学 | 一种提高视频空间分辨率的方法 |
CN108989731A (zh) * | 2018-08-09 | 2018-12-11 | 复旦大学 | 一种提高视频空间分辨率的方法 |
CN109741254A (zh) * | 2018-12-12 | 2019-05-10 | 深圳先进技术研究院 | 字典训练及图像超分辨重建方法、系统、设备及存储介质 |
CN109741254B (zh) * | 2018-12-12 | 2022-09-27 | 深圳先进技术研究院 | 字典训练及图像超分辨重建方法、系统、设备及存储介质 |
WO2021093620A1 (en) * | 2019-11-15 | 2021-05-20 | Huawei Technologies Co., Ltd. | Method and system for high-resolution image inpainting |
US11501415B2 (en) | 2019-11-15 | 2022-11-15 | Huawei Technologies Co. Ltd. | Method and system for high-resolution image inpainting |
CN111709442A (zh) * | 2020-05-07 | 2020-09-25 | 北京工业大学 | 一种面向图像分类任务的多层字典学习方法 |
CN111951167B (zh) * | 2020-08-25 | 2021-05-18 | 深圳思谋信息科技有限公司 | 超分辨率图像重建方法、装置、计算机设备和存储介质 |
CN111951167A (zh) * | 2020-08-25 | 2020-11-17 | 深圳思谋信息科技有限公司 | 超分辨率图像重建方法、装置、计算机设备和存储介质 |
WO2022087778A1 (zh) * | 2020-10-26 | 2022-05-05 | 深圳大学 | 一种基于多层耦合映射的低分辨率图像识别方法 |
CN113158928A (zh) * | 2021-04-27 | 2021-07-23 | 浙江云奕科技有限公司 | 一种基于图像识别的混凝土试块防造假方法 |
CN113158928B (zh) * | 2021-04-27 | 2023-09-19 | 浙江云奕科技有限公司 | 一种基于图像识别的混凝土试块防造假方法 |
CN114492450A (zh) * | 2021-12-22 | 2022-05-13 | 马上消费金融股份有限公司 | 文本匹配方法及装置 |
CN114972043A (zh) * | 2022-08-03 | 2022-08-30 | 江西财经大学 | 基于联合三边特征滤波的图像超分辨率重建方法与系统 |
CN114972043B (zh) * | 2022-08-03 | 2022-10-25 | 江西财经大学 | 基于联合三边特征滤波的图像超分辨率重建方法与系统 |
Also Published As
Publication number | Publication date |
---|---|
US20180232857A1 (en) | 2018-08-16 |
US10339633B2 (en) | 2019-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2017075768A1 (zh) | 一种基于字典匹配的图像超分辨率重建方法及装置 | |
CN112329800B (zh) | 一种基于全局信息引导残差注意力的显著性目标检测方法 | |
CN112750082B (zh) | 基于融合注意力机制的人脸超分辨率方法及系统 | |
CN110544205B (zh) | 基于可见光与红外交叉输入的图像超分辨率重建方法 | |
Zheng et al. | Residual multiscale based single image deraining | |
CN111340744B (zh) | 基于注意力双流深度网络的低质量图像降采样方法及其系统 | |
Huang et al. | Deep hyperspectral image fusion network with iterative spatio-spectral regularization | |
CN107220957B (zh) | 一种利用滚动导向滤波的遥感图像融合方法 | |
CN105389778B (zh) | 一种基于字典匹配的图像超分辨率重建方法及装置 | |
CN114266957B (zh) | 一种基于多降质方式数据增广的高光谱图像超分辨率复原方法 | |
WO2020043296A1 (en) | Device and method for separating a picture into foreground and background using deep learning | |
CN110738660A (zh) | 基于改进U-net的脊椎CT图像分割方法及装置 | |
CN114372962B (zh) | 基于双粒度时间卷积的腹腔镜手术阶段识别方法与系统 | |
WO2017070841A1 (zh) | 图像处理方法和装置 | |
WO2015180055A1 (zh) | 一种基于分类字典库的超分辨率图像重构方法及装置 | |
Senthilkani et al. | Overlap wavelet transform for image segmentation | |
Abed et al. | Architectural heritage images classification using deep learning with CNN | |
Hua et al. | Dynamic scene deblurring with continuous cross-layer attention transmission | |
CN108764287B (zh) | 基于深度学习和分组卷积的目标检测方法及系统 | |
Xu et al. | Degradation-aware dynamic fourier-based network for spectral compressive imaging | |
Jadhav | Image fusion based on wavelet transform | |
CN104504162A (zh) | 一种基于机器人视觉平台的视频检索方法 | |
CN115565034A (zh) | 基于双流增强网络的红外小目标检测方法 | |
CN114723948A (zh) | 基于分层次细节增强的高分辨率遥感图像语义分割方法 | |
Zhong et al. | An efficient saliency detection model based on wavelet generalized lifting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15907610 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15749554 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04/09/2018) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15907610 Country of ref document: EP Kind code of ref document: A1 |