CN105160627A - Method and system for super-resolution image acquisition based on classified self-learning - Google Patents
Method and system for super-resolution image acquisition based on classified self-learning Download PDFInfo
- Publication number
- CN105160627A CN105160627A CN201510552110.3A CN201510552110A CN105160627A CN 105160627 A CN105160627 A CN 105160627A CN 201510552110 A CN201510552110 A CN 201510552110A CN 105160627 A CN105160627 A CN 105160627A
- Authority
- CN
- China
- Prior art keywords
- resolution image
- low
- image
- super
- smoothness
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 76
- 238000005070 sampling Methods 0.000 claims abstract description 57
- 230000008878 coupling Effects 0.000 claims description 20
- 238000010168 coupling process Methods 0.000 claims description 20
- 238000005859 coupling reaction Methods 0.000 claims description 20
- 238000003384 imaging method Methods 0.000 claims description 19
- 238000005520 cutting process Methods 0.000 claims description 14
- 238000004513 sizing Methods 0.000 claims description 12
- 238000009499 grossing Methods 0.000 claims description 9
- 230000020509 sex determination Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 abstract 1
- 230000035939 shock Effects 0.000 abstract 1
- 230000009514 concussion Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 230000003321 amplification Effects 0.000 description 4
- 238000003199 nucleic acid amplification method Methods 0.000 description 4
- 230000007423 decrease Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 101100457838 Caenorhabditis elegans mod-1 gene Proteins 0.000 description 2
- 101150110972 ME1 gene Proteins 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000013011 mating Effects 0.000 description 1
- 238000013341 scale-up Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for super-resolution image acquisition based on classified self-learning. The method comprises the following steps of presetting and storing a smoothness judging window and a smoothness judging threshold; utilizing the smoothness judging window to carry out smoothness judgment on a low-resolution image; after sampling the low-resolution image twice, obtaining a high-frequency image, which is corresponding to the low-resolution image and only includes high-frequency information; and respectively adopting different matching methods according to a smoothness judging result of the low-resolution image, and carrying out high-frequency enhancement on the matched image to generate a super-resolution image. The method and the system respectively adopt different self-learning resolution processing methods for a smooth part and a sharp part of the low-resolution image, the image resolution is improved while the image noise and the edge shock are reduced, and the definition of the low-resolution image is improved.
Description
Technical field
The present invention relates to digital image processing field, particularly relate to a kind of super-resolution image acquisition method based on classification self study and system.
Background technology
Along with continuing to optimize of televisor display technique, the television screen generally used in general family constantly increases, and the requirement of people to television image pixel is more and more higher.The expression effect of image on large screen television of low resolution is poor, is more and more difficult to the viewing demand meeting people.The image of super-resolution can be obtained by special image procossing sensor, but the requirement of the imageing sensor of super-resolution to technique and cost is all higher, so application image Processing Algorithm obtains super-resolution image.
Current image super-resolution rebuilding technology is mainly divided into two kinds of patterns, and 1: according to one group of sequence of low resolution pictures of Same Scene, count the shifted relationship between image, by the different details using different images to provide, estimate a high-resolution image.2: according to the input of single low-resolution image, calculate the super-resolution image exported.
In first method, first need to carry out estimation accurately to sequence of low resolution pictures, thus registration is carried out to image, the image result of super-resolution is reconstructed afterwards again according to methods such as this regularization of series of low resolution image applications, maximum a posteriori probabilities.In this process, estimation and image registration are difficult to realize registration accurately, and error hiding can bring considerable influence to the reconstruction of follow-up super-resolution image.
And second method, single low-resolution image is processed, the process of estimation and image registration can be avoided, but because single low-resolution image source can only provide limited image information, so the super-resolution image effect that conventional interpolation amplification algorithm obtains is poor, often there is sawtooth or fuzzy phenomenon in the super-resolution rebuilding result obtained.Based on the single-frame images super-resolution rebuilding algorithm of self study, super-resolution rebuilding process can be carried out to single-frame images, and result images comparatively clearly can be obtained, but in actual process, edge concussion is easily produced to the image of the clear-cut margin such as word, concentric circles.
Therefore, prior art haves much room for improvement and improves.
Summary of the invention
In view of the deficiencies in the prior art, the object of the invention is to provide a kind of super-resolution image acquisition method based on classification self study and system, be intended to solve in prior art and cannot carry out registration accurately by low resolution figure to during image registration, when utilizing single low-resolution image to carry out image reconstruction, image border easily produces edge concussion, the ropy defect of the super-resolution image after reconstruction.
Technical scheme of the present invention is as follows:
Based on a super-resolution image acquisition method for classification self study, wherein, method comprises:
A, pre-set a smoothness decision window and smoothness decision threshold and store;
B, utilize smoothness decision window to the smoothing sex determination of low-resolution image;
C, double sampling is carried out to low-resolution image after obtain the high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image;
D, flatness result of determination according to low-resolution image, adopt different matching process respectively, and carry out high frequency enhancement to the image after coupling, generates super-resolution image.
The described super-resolution image acquisition method based on classification self study, wherein, described steps A specifically comprises:
The window ranges of A1, in advance a selection pre-sizing is as smoothness decision window, and in application window, pixel value standard deviation calculates standard as smoothness and stores;
A2, pre-set a smoothness decision threshold and store.
The described super-resolution image acquisition method based on classification self study, wherein, described step B specifically also comprises:
B1, to low-resolution image by the standard deviation of pixel value in the window ranges calculation window of pre-sizing, and whether criterion difference lower than the smoothness decision threshold pre-set;
If B2 standard deviation is lower than the smoothness decision threshold pre-set, then judge it is smooth region;
If B3 standard deviation is not less than the smoothness decision threshold pre-set, then judge it is cutting region.
The described super-resolution image acquisition method based on classification self study, wherein, described step C specifically comprises:
C1, to input need low-resolution image to be processed carry out down-sampling process, obtain the first sampled images;
C2, up-sampling process is carried out to the first sampled images, obtain the second sampled images;
C3, low-resolution image deducted the high frequency imaging that the second sampled images only comprised high-frequency information.
The described super-resolution image acquisition method based on classification self study, wherein, described step D specifically comprises:
D1, up-sampling process is carried out to low-resolution image, obtain the 3rd sampled images;
D2, be judged to be cutting region in low-resolution image, the 3rd sampled images is mated with low-resolution image;
D3, be judged to be smooth region in low-resolution image, the 3rd sampled images is mated with the second sampled images;
D4, utilize high frequency imaging to coupling after image strengthen, generate super-resolution image.
Super-resolution image based on classification self study obtains a system, and wherein, system comprises:
Pre-set and memory module, for pre-setting a smoothness decision window and smoothness decision threshold and storing;
Flatness determination module, for utilizing smoothness decision window to the smoothing sex determination of low-resolution image;
Sampling module, for obtaining the high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image after carrying out double sampling to low-resolution image;
Coupling and Computer image genration module, for the flatness result of determination according to low-resolution image, adopt different matching process respectively, and carry out high frequency enhancement to the image after coupling, generates super-resolution image.
The described super-resolution image based on classification self study obtains system, wherein, described in pre-set and specifically comprise with memory module:
First pre-sets and storage unit, and for selecting the window ranges of a pre-sizing as smoothness decision window in advance, in application window, pixel value standard deviation calculates standard as smoothness and stores;
Second pre-sets unit, for pre-setting a smoothness decision threshold and storing.
The described super-resolution image based on classification self study obtains system, and wherein, described flatness determination module specifically comprises:
Computing unit, for low-resolution image by the standard deviation of pixel value in the window ranges calculation window of pre-sizing, and whether criterion difference lower than the smoothness decision threshold pre-set;
First identifying unit, if for standard deviation lower than the smoothness decision threshold pre-set, then judges it is smooth region;
Second identifying unit, if be not less than the smoothness decision threshold pre-set for standard deviation, then judges it is cutting region.
The described super-resolution image based on classification self study obtains system, and wherein, described sampling module specifically comprises:
First sampling unit, for carrying out down-sampling process to the need low-resolution image to be processed of input, obtains the first sampled images;
Second sampling unit, for carrying out up-sampling process to the first sampled images, obtains the second sampled images;
Graphics processing unit, for deducting the high frequency imaging that the second sampled images is only comprised high-frequency information by low-resolution image.
The described super-resolution image based on classification self study obtains system, and wherein, described sampling module specifically comprises:
3rd sampling unit, carries out up-sampling process to low-resolution image, obtains the 3rd sampled images;
First image matching unit, for being judged to be cutting region in low-resolution image, mates the 3rd sampled images with low-resolution image;
Second image matching unit, for being judged to be smooth region in low-resolution image, mates the 3rd sampled images with the second sampled images;
Image enhancing unit, for utilizing high frequency imaging to strengthen the image after coupling, generates super-resolution image.
Beneficial effect: first the image that the present invention is low to resolution has carried out the judgement of smooth and piercing portion, adopts different image samplings and matching process respectively after judgement for different piece, and utilizes the high-frequency information in image to carry out image enhaucament.The present invention, while raising image resolution ratio, decreases picture noise and edge concussion, improves the sharpness of low-resolution image, obtain the super-resolution image of edge-smoothing.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the preferred embodiment of this kind of super-resolution image acquisition method based on classification self study of the present invention.
Fig. 2 is that the smoothness of the embody rule embodiment of a kind of super-resolution image acquisition method based on classification self study of the present invention judges schematic diagram.
Fig. 3 is the image sampling schematic diagram of the embody rule embodiment of a kind of super-resolution image acquisition method based on classification self study of the present invention.
Fig. 4 is the super-resolution image reconstruction schematic diagram of the embody rule embodiment of a kind of super-resolution image acquisition method based on classification self study of the present invention.
Fig. 5 is the functional schematic block diagram that a kind of super-resolution image based on classification self study of the present invention obtains the preferred embodiment of system.
Embodiment
For making object of the present invention, technical scheme and effect clearly, clearly, the present invention is described in more detail below.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.
The invention provides a kind of process flow diagram of preferred embodiment of the super-resolution image acquisition method based on classification self study, as shown in Figure 1, described method comprises:
Step S100, pre-set a smoothness decision window and smoothness decision threshold and store.
Further, described step S100 specifically comprises:
The window ranges of step S101, in advance a selection pre-sizing is as smoothness decision window, and in application window, pixel value standard deviation calculates standard as smoothness and stores;
Step S102, pre-set a smoothness decision threshold and store.
During concrete enforcement, in the present invention, pre-set a smoothness decision window, preferably neighborhood territory pixel is got within the scope of its 5X5 as smoothness decision window, to pixel value standard deviation in decision window as smoothness criterion to each pixel.Be designated as s for the pixel value standard deviation within the scope of the 5X5 of each pixel, the pixel value at this pixel place is designated as I (x, y), then in window, the computing formula of pixel value standard deviation is:
Smoothness decision threshold can need to arrange according to user, and smoothness is judged that thresholding is set to 5 by preferred the present invention.Lower than the part of decision threshold, be set to smooth region; Higher than the part of decision threshold, be set to cutting region.
Step S200, utilize smoothness decision window to the smoothing sex determination of low-resolution image.
Further, described step S200 specifically comprises:
Step S201, to low-resolution image by the standard deviation of pixel value in the window ranges calculation window of pre-sizing, and whether criterion difference lower than the smoothness decision threshold pre-set;
If step S202 standard deviation is lower than the smoothness decision threshold pre-set, then judge it is smooth region;
If step S203 standard deviation is not less than the smoothness decision threshold pre-set, then judge it is cutting region.
During concrete enforcement, each pixel in low-resolution image is got it and judges according to the decision window pre-set.Particularly its 5X5 neighborhood is got to each pixel in low-resolution image interval, calculate each pixel window internal standard poor.Setting smoothness decision threshold.Standard deviation is greater than the pixel of decision threshold, adjudicates as cutting region, and its smoothness template value is 1.Standard deviation is less than to the pixel of decision threshold, adjudicate it for smooth region, in smoothness template, value is 0.
After obtaining the two-value smoothness judgement template of low-resolution image, two-value template is amplified, obtain the smoothness judgement template of super-resolution image.Because smoothness judgement template is the same with the size of super-resolution image, be obtain after low-resolution image amplifies.Need below, according to smoothness template, to judge the match pattern that in super-resolution image, each location of pixels is corresponding.
The structural scheme of super-resolution image is divided into two parts by smoothness judgement template, and value is 1 be judged as mod1, and value is 0 be judged as mod0.As shown in Figure 2, in figure, value is the mod1 of 1 is the first template, value be 0 mod0 be the second template, first template is the intersection of black and white in Fig. 2, image intensity value changes greatly, and the second template is that image intensity value changes less part, as the region of black in Fig. 2.
Step S300, double sampling is carried out to low-resolution image after obtain the high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image.
Further, described step S300 specifically comprises:
Step S301, to input need low-resolution image to be processed carry out down-sampling process, obtain the first sampled images;
Step S302, up-sampling process is carried out to the first sampled images, obtain the second sampled images;
Step S303, low-resolution image deducted the high frequency imaging that the second sampled images only comprised high-frequency information.
In acquisition low frequency part image process, first need the low-frequency image L0 obtaining more small scale.Wherein, image down sampling reduces the scale down 2/3 of use.This ratio can be chosen by demand.As shown in Figure 3.Coordinate relation between the low-resolution image L of current input and low yardstick low-frequency image L0, can predict the coordinate relation between the high-definition picture L_out of final acquisition and low-resolution image L at the corresponding levels.Pointwise coupling is carried out to current low-resolution image L and L0, the point (the solid stain in Fig. 3 in L0) finding each pixel of low-resolution image L of input to match in L0 and the distance r of its coordinate in proportion between mapping center point (hollow dots in Fig. 3 in L0).In the process of super-resolution rebuilding L_out, each element in the region that in L, each pixel is corresponding in L_out (in Fig. 3 in L_out in white portion each pixel), in the process of mating with L1, window size be chosen as r', r' equals r and is multiplied by magnification ratio.Picture up-sampling amplifies the magnification ratio 3/2 used, and this ratio can be chosen by demand, and the up-sampling in the embodiment of the present invention amplifies and down-sampling scale down is consistent.Namely after original image first reduces rear amplification, size will remain unchanged.
During concrete enforcement, as Fig. 3, carry out downsampled to low-resolution image, also known as downsample, obtain the low-resolution image L0 of more small scale.Pointwise coupling is carried out to L and L0.The point (the solid stain in Fig. 3 in L0) that each pixel recording L matches in L0 and the distance of its coordinate in proportion between mapping center point (hollow dots in Fig. 3 in L0).Obtain the search window radius template of low-resolution image.Scale up, obtain the search window radius of super-resolution result.Picture up-sampling amplifies the magnification ratio 3/2 used, and this ratio can be chosen by demand, and the scale down of up-sampling amplification herein and before down-sampling is consistent.
Down-sampling is carried out to the low-resolution image L0 of input, obtains down-sampled images L0, up-sampling is carried out to L0, obtain the low frequency part L1 of input picture.Comprise the low-frequency information part of former figure in L1, high frequency detail can obtain H1 by deducting L1 with former figure L, i.e. H1 in Fig. 4.H1 characterizes the detail of the high frequency of image.
Step S400, flatness result of determination according to low-resolution image, adopt different matching process respectively, and carry out high frequency enhancement to the image after coupling, generates super-resolution image.
Further, described step S400 specifically comprises:
Step S401, up-sampling process is carried out to low-resolution image, obtain the 3rd sampled images;
Step S402, be judged to be cutting region in low-resolution image, the 3rd sampled images is mated with low-resolution image;
Step S403, be judged to be smooth region in low-resolution image, the 3rd sampled images is mated with the second sampled images;
Step S404, utilize high frequency imaging to coupling after image strengthen, generate super-resolution image.
During concrete enforcement, directly up-sampling is carried out to the low-resolution image L of input, obtain the low frequency part L2 of super-resolution image.L2 finds the correspondence position of each point in former low-resolution image in L2 according to a preliminary estimate to one of super-resolution image result, and is filled in L2 by the high-frequency information in L2 corresponding for this position and goes, and just can obtain final super-resolution result.As shown in Figure 3.
First, according to smoothness template, judge the match pattern that in L2, each location of pixels is corresponding, if this pixel is in smooth area, corresponding mode2 pattern, so applies former figure L and mates with it, if this pixel is in sharp keen district, corresponding mode1 pattern, the low frequency part L1 so applying former figure mates with it.Then according to the search window radius of prediction, obtain final matched position, thus choose the high-frequency information of matched position in H1 L2 is strengthened, finally obtain complete super-resolution and amplify result.
According to smoothness template and the match window yardstick estimated, to the low-frequency image L1 after amplification according to different smoothnesses, different match objects and matching scheme is selected to mate, after retrieving optimum matching, corresponding HFS is added on L1, thus gets the super-resolution image result of non-flanged concussion and noise.
As shown in Figure 4, final super-resolution image L_out is that the high-frequency information H1 by adding former figure matched position in the low-frequency image L2 amplified obtains, so, find correct matched position most important.If the high-frequency information chosen too is dispersed, the final image result obtained can be level and smooth, but edge easily produces concussion and fuzzy; If it is too concentrated to choose high frequency position, image is stiff excessive by occurring, not smoothly.
Because low-resolution image L is compared with its low frequency part L1, have details clear, the features such as unique point is obvious, in the process of coupling, it can provide locates more accurately, makes the selection of HFS more concentrate on texture part.So for part sharp keen in smooth template, the part that namely texture is more, this method is chosen the former figure L of low resolution and is mated.
And do not comprise high-frequency information due to the low frequency part L1 of low-resolution image itself, be in comparatively level and smooth level, have between neighbor and certain seamlessly transit phenomenon, mate with L1, obtain than being easier to the high-frequency information dispersed, the integral image obtained is in comparatively level and smooth level.So, for the pixel of smooth in smooth template, choose low-frequency image L1 and mate, less noise can be introduced, eliminate stiff transient, obtain level and smooth image result.
Matching criterior chooses SAD absolute error criterion:
From above embodiment of the method, the invention provides a kind of super-resolution image acquisition method based on classification self study, by first having carried out the judgement of smooth and piercing portion to low-resolution image, by obtaining the high-frequency information image in image to image sampling, and adopt Different matching method respectively to the result that low-resolution image judges according to flatness, and the high-frequency information in image is utilized to carry out image enhaucament to the image after coupling.The present invention, while raising image resolution ratio, decreases picture noise and edge concussion, improves the sharpness of low-resolution image, obtain the super-resolution image of edge-smoothing.
On the basis of said method embodiment, present invention also offers the functional schematic block diagram that a kind of super-resolution image based on classification self study obtains the preferred embodiment of system, as shown in Figure 5, described system comprises:
Pre-set and memory module 100, for pre-setting a smoothness decision window and smoothness decision threshold and storing; As detailed above.
Flatness determination module 200, for utilizing smoothness decision window to the smoothing sex determination of low-resolution image; As detailed above.
Sampling module 300, for obtaining the high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image after carrying out double sampling to low-resolution image; As detailed above.
Coupling and Computer image genration module 400, for the flatness result of determination according to low-resolution image, adopt different matching process respectively, and carry out high frequency enhancement to the image after coupling, generates super-resolution image; As detailed above.
The described super-resolution image based on classification self study obtains system, wherein, described in pre-set and specifically comprise with memory module:
First pre-sets and storage unit, and for selecting the window ranges of a pre-sizing as smoothness decision window in advance, in application window, pixel value standard deviation calculates standard as smoothness and stores; As detailed above.
Second pre-sets unit, for pre-setting a smoothness decision threshold and storing; As detailed above.
The described super-resolution image based on classification self study obtains system, and wherein, described flatness determination module specifically comprises:
Computing unit, for low-resolution image by the standard deviation of pixel value in the window ranges calculation window of pre-sizing, and whether criterion difference lower than the smoothness decision threshold pre-set; As detailed above.
First identifying unit, if for standard deviation lower than the smoothness decision threshold pre-set, then judges it is smooth region; As detailed above.
Second identifying unit, if be not less than the smoothness decision threshold pre-set for standard deviation, then judges it is cutting region; As detailed above.
The described super-resolution image based on classification self study obtains system, and wherein, described sampling module specifically comprises:
First sampling unit, for carrying out down-sampling process to the need low-resolution image to be processed of input, obtains the first sampled images; As detailed above.
Second sampling unit, for carrying out up-sampling process to the first sampled images, obtains the second sampled images; As detailed above.
Graphics processing unit, for deducting the high frequency imaging that the second sampled images is only comprised high-frequency information by low-resolution image; As detailed above.
The described super-resolution image based on classification self study obtains system, and wherein, described sampling module specifically comprises:
3rd sampling unit, carries out up-sampling process to low-resolution image, obtains the 3rd sampled images; As detailed above.
First image matching unit, for being judged to be cutting region in low-resolution image, mates the 3rd sampled images with low-resolution image; As detailed above.
Second image matching unit, for being judged to be smooth region in low-resolution image, mates the 3rd sampled images with the second sampled images; As detailed above.
Image enhancing unit, for utilizing high frequency imaging to strengthen the image after coupling, generates super-resolution image; As detailed above.
In sum, the invention provides a kind of based on the classification super-resolution image acquisition method of self study and system, described method comprises: pre-set a smoothness decision window and smoothness decision threshold and store; Utilize smoothness decision window to the smoothing sex determination of low-resolution image; The high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image is obtained after double sampling is carried out to low-resolution image; According to the flatness result of determination of low-resolution image, adopt different matching process respectively, and high frequency enhancement is carried out to the image after coupling, generate super-resolution image.The present invention can adopt the resolution processes method of different self studies respectively to the smooth of the low image of resolution and piercing portion, while raising image resolution ratio, decrease picture noise and edge concussion, improve the sharpness of low-resolution image.
Should be understood that, application of the present invention is not limited to above-mentioned citing, for those of ordinary skills, can be improved according to the above description or convert, and all these improve and convert the protection domain that all should belong to claims of the present invention.
Claims (10)
1., based on a super-resolution image acquisition method for classification self study, it is characterized in that, method comprises:
A, pre-set a smoothness decision window and smoothness decision threshold and store;
B, utilize smoothness decision window to the smoothing sex determination of low-resolution image;
C, double sampling is carried out to low-resolution image after obtain the high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image;
D, flatness result of determination according to low-resolution image, adopt different matching process respectively, and carry out high frequency enhancement to the image after coupling, generates super-resolution image.
2., according to claim 1 based on the super-resolution image acquisition method of classification self study, it is characterized in that, described steps A specifically comprises:
The window ranges of A1, in advance a selection pre-sizing is as smoothness decision window, and in application window, pixel value standard deviation calculates standard as smoothness and stores;
A2, pre-set a smoothness decision threshold and store.
3., according to claim 2 based on the super-resolution image acquisition method of classification self study, it is characterized in that, described step B specifically also comprises:
B1, to low-resolution image by the standard deviation of pixel value in the window ranges calculation window of pre-sizing, and whether criterion difference lower than the smoothness decision threshold pre-set;
If B2 standard deviation is lower than the smoothness decision threshold pre-set, then judge it is smooth region;
If B3 standard deviation is not less than the smoothness decision threshold pre-set, then judge it is cutting region.
4., according to claim 3 based on the super-resolution image acquisition method of classification self study, it is characterized in that, described step C specifically comprises:
C1, to input need low-resolution image to be processed carry out down-sampling process, obtain the first sampled images;
C2, up-sampling process is carried out to the first sampled images, obtain the second sampled images;
C3, low-resolution image deducted the high frequency imaging that the second sampled images only comprised high-frequency information.
5., according to claim 4 based on the super-resolution image acquisition method of classification self study, it is characterized in that, described step D specifically comprises:
D1, up-sampling process is carried out to low-resolution image, obtain the 3rd sampled images;
D2, be judged to be cutting region in low-resolution image, the 3rd sampled images is mated with low-resolution image;
D3, be judged to be smooth region in low-resolution image, the 3rd sampled images is mated with the second sampled images;
D4, utilize high frequency imaging to coupling after image strengthen, generate super-resolution image.
6. the super-resolution image based on classification self study obtains a system, and it is characterized in that, system comprises:
Pre-set and memory module, for pre-setting a smoothness decision window and smoothness decision threshold and storing;
Flatness determination module, for utilizing smoothness decision window to the smoothing sex determination of low-resolution image;
Sampling module, for obtaining the high frequency imaging only comprising high-frequency information of the correspondence of low-resolution image after carrying out double sampling to low-resolution image;
Coupling and Computer image genration module, for the flatness result of determination according to low-resolution image, adopt different matching process respectively, and carry out high frequency enhancement to the image after coupling, generates super-resolution image.
7. obtain system based on the super-resolution image of classification self study according to claim 6, it is characterized in that, described in pre-set and specifically comprise with memory module:
First pre-sets and storage unit, and for selecting the window ranges of a pre-sizing as smoothness decision window in advance, in application window, pixel value standard deviation calculates standard as smoothness and stores;
Second pre-sets unit, for pre-setting a smoothness decision threshold and storing.
8. obtain system based on the super-resolution image of classification self study according to claim 7, it is characterized in that, described flatness determination module specifically comprises:
Computing unit, for low-resolution image by the standard deviation of pixel value in the window ranges calculation window of pre-sizing, and whether criterion difference lower than the smoothness decision threshold pre-set;
First identifying unit, if for standard deviation lower than the smoothness decision threshold pre-set, then judges it is smooth region;
Second identifying unit, if be not less than the smoothness decision threshold pre-set for standard deviation, then judges it is cutting region.
9. obtain system based on the super-resolution image of classification self study according to claim 8, it is characterized in that, described sampling module specifically comprises:
First sampling unit, for carrying out down-sampling process to the need low-resolution image to be processed of input, obtains the first sampled images;
Second sampling unit, for carrying out up-sampling process to the first sampled images, obtains the second sampled images;
Graphics processing unit, for deducting the high frequency imaging that the second sampled images is only comprised high-frequency information by low-resolution image.
10. obtain system based on the super-resolution image of classification self study according to claim 9, it is characterized in that, described sampling module specifically comprises:
3rd sampling unit, carries out up-sampling process to low-resolution image, obtains the 3rd sampled images;
First image matching unit, for being judged to be cutting region in low-resolution image, mates the 3rd sampled images with low-resolution image;
Second image matching unit, for being judged to be smooth region in low-resolution image, mates the 3rd sampled images with the second sampled images;
Image enhancing unit, for utilizing high frequency imaging to strengthen the image after coupling, generates super-resolution image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510552110.3A CN105160627B (en) | 2015-08-31 | 2015-08-31 | Super-resolution image acquisition method and system based on classification self-learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510552110.3A CN105160627B (en) | 2015-08-31 | 2015-08-31 | Super-resolution image acquisition method and system based on classification self-learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105160627A true CN105160627A (en) | 2015-12-16 |
| CN105160627B CN105160627B (en) | 2020-06-23 |
Family
ID=54801470
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510552110.3A Active CN105160627B (en) | 2015-08-31 | 2015-08-31 | Super-resolution image acquisition method and system based on classification self-learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105160627B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108714003A (en) * | 2018-05-27 | 2018-10-30 | 浠诲嘲 | The fixation desktop maintenance system measured based on gray scale |
| CN108932694A (en) * | 2017-05-26 | 2018-12-04 | 深圳赛奥航空科技有限公司 | A kind of mould group optimizing image resolution ratio |
| CN111383299A (en) * | 2018-12-28 | 2020-07-07 | Tcl集团股份有限公司 | Image processing method and device and computer readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007122838A1 (en) * | 2006-04-25 | 2007-11-01 | National University Corporation NARA Institute of Science and Technology | Super-resolution method based on hierarchy bayes approach and super-resolution program |
| CN101877124A (en) * | 2009-04-28 | 2010-11-03 | 北京捷科惠康科技有限公司 | Method and system for filtering medical image |
| CN102378011A (en) * | 2010-08-12 | 2012-03-14 | 华为技术有限公司 | Method, device and system for up-sampling image |
| CN103475876A (en) * | 2013-08-27 | 2013-12-25 | 北京工业大学 | Learning-based low-bit-rate compression image super-resolution reconstruction method |
| CN103489174A (en) * | 2013-10-08 | 2014-01-01 | 武汉大学 | Human face super-resolution method based on residual keeping |
| CN103903236A (en) * | 2014-03-10 | 2014-07-02 | 北京信息科技大学 | Method and device for reconstructing super-resolution facial image |
-
2015
- 2015-08-31 CN CN201510552110.3A patent/CN105160627B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007122838A1 (en) * | 2006-04-25 | 2007-11-01 | National University Corporation NARA Institute of Science and Technology | Super-resolution method based on hierarchy bayes approach and super-resolution program |
| CN101877124A (en) * | 2009-04-28 | 2010-11-03 | 北京捷科惠康科技有限公司 | Method and system for filtering medical image |
| CN102378011A (en) * | 2010-08-12 | 2012-03-14 | 华为技术有限公司 | Method, device and system for up-sampling image |
| CN103475876A (en) * | 2013-08-27 | 2013-12-25 | 北京工业大学 | Learning-based low-bit-rate compression image super-resolution reconstruction method |
| CN103489174A (en) * | 2013-10-08 | 2014-01-01 | 武汉大学 | Human face super-resolution method based on residual keeping |
| CN103903236A (en) * | 2014-03-10 | 2014-07-02 | 北京信息科技大学 | Method and device for reconstructing super-resolution facial image |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108932694A (en) * | 2017-05-26 | 2018-12-04 | 深圳赛奥航空科技有限公司 | A kind of mould group optimizing image resolution ratio |
| CN108714003A (en) * | 2018-05-27 | 2018-10-30 | 浠诲嘲 | The fixation desktop maintenance system measured based on gray scale |
| CN111383299A (en) * | 2018-12-28 | 2020-07-07 | Tcl集团股份有限公司 | Image processing method and device and computer readable storage medium |
| CN111383299B (en) * | 2018-12-28 | 2022-09-06 | Tcl科技集团股份有限公司 | Image processing method and device and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105160627B (en) | 2020-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9639918B2 (en) | Method for anti-aliasing of image with super-resolution | |
| Wang et al. | Dehazing for images with large sky region | |
| CN106651938A (en) | Depth map enhancement method blending high-resolution color image | |
| CN110634147B (en) | Image matting method based on bilateral guide up-sampling | |
| CN101976436B (en) | Pixel-level multi-focus image fusion method based on difference image correction | |
| CN103632359A (en) | Super-resolution processing method for videos | |
| Hua et al. | Extended guided filtering for depth map upsampling | |
| CN109214380B (en) | License plate inclination correction method | |
| CN104376551A (en) | Color image segmentation method integrating region growth and edge detection | |
| US20130170736A1 (en) | Disparity estimation depth generation method | |
| CN106204441B (en) | Image local amplification method and device | |
| CN104574366A (en) | Extraction method of visual saliency area based on monocular depth map | |
| CN105139790B (en) | OLED shows aging method for detecting and display device | |
| CN104574328A (en) | Color image enhancement method based on histogram segmentation | |
| CN116797590B (en) | Mura defect detection method and system based on machine vision | |
| CN103914820A (en) | Image haze removal method and system based on image layer enhancement | |
| CN102289786B (en) | Edge anti-aliasing method and device for image scaling | |
| CN105139338A (en) | Multi-dimensional lookup table generation method and device and image scaling processing method and device | |
| CN102930515A (en) | Automatic geometric distortion correction method of digital image | |
| CN113221925A (en) | Target detection method and device based on multi-scale image | |
| CN103826032A (en) | Depth map post-processing method | |
| CN105160627A (en) | Method and system for super-resolution image acquisition based on classified self-learning | |
| CN105354795A (en) | Phase correlation based acquisition method and system for self-learning super-resolution image | |
| CN102903111B (en) | Large area based on Iamge Segmentation low texture area Stereo Matching Algorithm | |
| CN104599288A (en) | Skin color template based feature tracking method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 516006 TCL technology building, No.17, Huifeng Third Road, Zhongkai high tech Zone, Huizhou City, Guangdong Province Applicant after: TCL Technology Group Co., Ltd Address before: 516006 Guangdong province Huizhou Zhongkai hi tech Development Zone No. nineteen District Applicant before: TCL RESEARCH AMERICA Inc. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |