CN105469373A - Retinex image enhancing method and system - Google Patents
Retinex image enhancing method and system Download PDFInfo
- Publication number
- CN105469373A CN105469373A CN201410450328.3A CN201410450328A CN105469373A CN 105469373 A CN105469373 A CN 105469373A CN 201410450328 A CN201410450328 A CN 201410450328A CN 105469373 A CN105469373 A CN 105469373A
- Authority
- CN
- China
- Prior art keywords
- image
- component
- module
- gray
- reflecting component
- 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
Landscapes
- Image Processing (AREA)
Abstract
The invention provides a Retinex image enhancing method and a Retinex image enhancing system. The Retinex image enhancing method comprises the steps of: carrying out repeated average filtering iteration on a video source image acquired by a video camera to generate an image illuminance component; acquiring an image reflection component according to the source image and the image illuminance component; correcting the image illuminance component and the image reflection component; and calculating to obtain an image after Retinex image enhancement according to the corrected image illuminance component and the corrected image reflection component. The Retinex image enhancing system comprises an iterative convolution module, a reflection component generation module, an illuminance component correction module, a reflection component correction module and an enhanced image generation module, wherein the modules are implemented on an FPGA. The Retinex image enhancing method and the Retinex image enhancing system can be used for carrying out self-adaptive enhancement on images of different types, and ensure the real-time performance of image enhancement.
Description
Technical field
The invention belongs to digital video image processing technology field, be specifically related to a kind of Retinex image enchancing method and system.
Background technology
Image enhaucament refers to by specifically needing to adopt ad hoc approach to give prominence to some information in image, weakens simultaneously or removes irrelevant information, or source images is converted to the image processing method that a kind of people of being more suitable for or machine carry out analyzing and processing form.Traditional images Enhancement Method mainly comprises image gray levels conversion, histogram equalization, Gamma correction, image sharpening, edge enhancing etc.It is more single that traditional images strengthens algorithm general utility functions, can only strengthen some local feature of a certain types of image or image, and dissimilar image needs constantly adjustment parameter to obtain strengthens effect preferably.
Retinex image enchancing method is a kind of relatively optimum method proposed for above-mentioned situation.Existing Retinex image enchancing method is as follows:
The product of the basic assumption of Retinex image enchancing method to be source images F (x, y) be image illumination component I (x, y) and image reflecting component R (x, y), as shown in formula (l),
F(x,y)＝R(x,y)×I(x,y)(1)
Wherein, x, y are image transverse and longitudinal coordinate points, and image illumination component I (x, y) represents surround lighting, determine the dynamic range of source images F (x, y); Image reflecting component R (x, y) represents the reflectivity properties of object, determines the image detail of source images F (x, y).
Image illumination component I (x, y) as shown in formula (2),
I(x,y)＝F(x,y)*G(x,y)(2)
G (x, y), for low pass convolution is around function, generally adopts Gaussian form, is used for from estimation image illumination component I (x, y) in source images F (x, y), its expression formula for shown in formula (3),
Wherein λ is scalar matrix, and it makes
∫∫G(x,y)dxdy＝1(4)
C is yardstick constant.
Then image reflecting component R (x, y) can represent as shown in formula (5),
R(x,y)＝F(x,y)/[F(x,y)*G(x,y)](5)
Retinex image enchancing method ultimate principle is exactly source images F (x, y) divided by source images F (x, y) and the convolution results of low pass convolution around function G (x, y), its objective is the impact eliminating uneven illumination, reach the object improving image visual effect.And low pass convolution is more sharp-pointed around function G (x, y), image reflecting component R (x, y) is more outstanding; Otherwise low pass convolution is more smooth around function G (x, y), then image illumination component I (x, y) keeps better.The method not only strengthens a certain category feature of image, can also reach balance, thus adaptively can strengthen dissimilar image in brightness reproduction, dynamic range compression and color constancy three.But Retinex image enchancing method adopts list/multiple dimensioned mask convolution computing, and computation complexity often exceeds the needs of practical application.When the method realizes on software platform, real-time is poor; Then take a large amount of internal memory by multi-core DSP hardware implementing, increase image processing system cost, real-time is not thoroughly improved simultaneously.
Summary of the invention
The technical matters that the present invention solves provides a kind of Retinex image enchancing method and system, and the method can carry out self-adaptation enhancing for dissimilar image, this system ensure that the real-time of image enhaucament.
In order to solve the problems of the technologies described above, the invention provides a kind of Retinex image enchancing method, comprising the following steps:
Step one, repeatedly mean filter grey iterative generation image illumination component is carried out to the video source image of camera acquisition, according to formula n=m/N, computation of mean values filtering iteration frequency n, wherein N is mean filter window width, and m is the width of source images;
Step 2, obtain image reflecting component according to source images and image illumination component;
Step 3, the correction of image illumination component: use statistics with histogram method, according to the pixel number of each gray level of image illumination component, synthetic image irradiates the histogram of component; All gray-level pixels points are added up, accumulation result is divided equally in all gray levels, generate histogram equalization mapping table; By histogram equalization mapping table, generate the image illumination component after histogram equalization;
Step 4, the correction of image reflecting component: use statistics with histogram method, according to the pixel number of each gray level of image reflecting component, the histogram of synthetic image reflecting component; According to pixel quantity threshold, read all gray-level pixels point and add up, calculate Adaptive Gray-Level scope and map gray level; By self organizing maps gray level, generate the image reflecting component after adaptive line conversion;
Step 5, according to revised image illumination component and image reflecting component, calculate and obtain image after Retinex image enhaucament.
The present invention also provides a kind of Retinex Image Intensified System, and comprise iterative convolution module, reflecting component generation module, luminance component correcting module, reflecting component correcting module, strengthen Computer image genration module, above-mentioned module all realizes on FPGA;
Iterative convolution module output terminal is connected with the input end of reflecting component generation module with luminance component correcting module simultaneously, the output terminal of reflecting component generation module is connected with the input end of reflecting component correcting module, the output terminal of reflecting component correcting module is connected with the input end strengthening Computer image genration module, and the output terminal of luminance component correcting module is connected with the input end strengthening Computer image genration module.
The video source image feeding iterative convolution module of camera acquisition is carried out repeatedly mean filter grey iterative generation luminance component and is sent to reflecting component generation module and luminance component correcting module;
After reflecting component generation module receives luminance component, read computational reflect component after the source images being stored in first piece of random access memory, then reflecting component is sent to reflecting component correcting module;
After reflecting component correcting module receives reflecting component, utilize adaptive gray-level transform to revise reflecting component, revised reflecting component sends to and strengthens Computer image genration module;
After luminance component correcting module receives luminance component, utilize histogram equalization to revise luminance component, revised luminance component sends to and strengthens Computer image genration module;
Strengthen Computer image genration module and receive revised reflecting component and luminance component, generate the image after strengthening.
In Retinex Image Intensified System of the present invention, the middle partial data produced utilizes block random access memory in sheet to carry out buffer memory.Be specially: between the video source image of camera acquisition and reflecting component generation module, be connected with first piece of random access memory, reflecting component generation module calls first piece of random access memory, luminance component correcting module intrinsic call second piece of random access memory and the 3rd piece of random access memory, reflecting component generation module intrinsic call the 4th piece of random access memory and the 5th piece of random access memory.The video source image of camera acquisition uses first piece of random access memory to carry out buffer memory; The gray-scale value number that luminance component correcting module carries out statistics with histogram is buffered in second piece of random access memory, and histogram equalization mapping table is buffered in the 3rd piece of random access memory; The gray-scale value number that reflecting component correcting module carries out self-adaptation gray scale stretching statistics is buffered in the 4th piece of random access memory, and self-adaptation gray scale stretching mapping table is buffered in the 5th piece of random access memory.
The present invention is to the improvement of low pass convolution around function
As shown in background technology Chinese style (3), the low pass convolution of Gaussian form is around function G (x, y), owing to there is exponential function, therefore calculation of complex during hardware implementing, the present invention selects mean filter to replace the low pass convolution of Gaussian form to carry out convolution around function G (x, y) to source images F (x, y).
Such as, for window size N=3 to source images signal
Mean filter is adopted to carry out convolution, a
i,jfor the gray-scale value that image coordinate is corresponding, m, n are that source images is long and wide.
Then after a mean filter, picture signal is
B
i,jit is the gray-scale value that after first time mean filter, image coordinate is corresponding.
Known by Mean Filtering Algorithm, any position b
i,jthe value of (1≤i≤n, 1≤j≤m) is a
i-1, j-1to a
i+1, j+1the average of continuous 9 numbers, namely
If the picture signal after second time mean filter
Wherein
c
ijit is the gray-scale value that after second time mean filter, image coordinate is corresponding.Work as i=1, during n, c
ij=b
ij.Launch c
ijobtain shown in formula (6)
In like manner, the picture signal after three mean filters
Wherein d
ijshown in (7), d
ijit is the gray-scale value that after third time mean filter, image coordinate is corresponding.
Equally, the iterative process of more times mean filter can be derived.Formula (6), (7) show, repeatedly picture signal arbitrfary point d after mean filter
ij(or c
ij) value depend on i, the weighted mean value of upper and lower data before and after j position, and weight data distribution possess Gaussian characteristics, namely meet the distribution character of formula (3).Therefore, repeatedly mean filter iteration can approach the low pass convolution of Gaussian form well around function G (x, y).
The correction of image illumination component of the present invention and image reflecting component
Image illumination component I (x, y) represents surround lighting, correspond to the low frequency component of image, determines the dynamic range that in source images F (x, y), pixel energy arrives, and has nothing to do with object.The present invention, by histogram equalization correction image luminance component I (x, y), improves the global contrast of image.Image reflecting component R (x, y) represents the reflectivity properties of object, correspond to the high fdrequency component in image, determines the image detail of source images F (x, y).Therefore, local detail can be strengthened by the value of adjustment image reflecting component R (x, y).
The principle of histogram equalization is, the gray scale of occupying more pixel increases with the extreme difference of previous gray level after the conversion, and target and background often occupies more pixel, thus increases the contrast of object and background; Meanwhile, the extreme difference of gray scale after the conversion with previous occupying less pixel is less, need merger, and the transition position pixel of border and background is less, due to merger, itself or become background dot or become impact point, thus make border become precipitous.Utilize histogram equalization to revise image illumination component I (x, y), thus improve the global contrast of image.
Self-adaptation gray scale stretches and mainly improves the dynamic range of image gray levels, and it selectively stretches between certain section of gray area to improve output image.If the gray scale of piece image concentrates on darker region and causes image partially dark, can stretch by gray scale stretch function between (slope >1) object gray area to improve image by gray scale; If same gradation of image concentrates on brighter region and causes image partially bright, also can compress by gray scale stretch function between (slope <1) object gray area to improve picture quality.Utilize self-adaptation gray scale to stretch to revise image reflecting component R (x, y), thus improve the local detail of image.
Through to above-mentioned to the adjustment of image illumination component and image reflecting component after, the overall situation and the local contrast enhancement processing of image can be realized.
The present invention compared with prior art, its remarkable advantage is, (1) the present invention adopts repeatedly mean filter iteration to avoid the low pass convolution of complicated multiple dimensioned Gaussian form around function G (x, y) mask convolution computing, by selecting rational scale size and iterations to ensure accuracy requirement, reduce the complexity of computing; (2) the present invention improves the local visibility of global contrast and details to image illumination component I (x, y) and the correction of image reflecting component R (x, y), and image enhancement effects is more obvious; (3) in the present invention, adopt block ram cache technology to carry out the view data produced in the middle of storage algorithm, utilize block random access memory to belong to FPGA on-chip memory, read-write operation is easier to, and circuit is simple, improves computing velocity; (4) Retinex Image Intensified System of the present invention adopts pure hardware implementing, so data transmission rate is fast, fast operation, real-time is good, and efficiency is high.
Accompanying drawing explanation
Fig. 1 is the structural representation of Retinex Image Intensified System of the present invention.
Fig. 2 is iterative convolution schematic diagram in the present invention.
Fig. 3 is luminance component correction schematic diagram in the present invention.
Fig. 4 is the histogram equalization process schematic diagram used in luminance component makeover process of the present invention.
Fig. 5 is the result figure that in the embodiment of the present invention, target scene uses the inventive method to obtain, wherein (a) is source images, b () is image illumination component, (c) image reflecting component, the image after (d) enhancing.
Embodiment
As shown in Figure 1, the present invention also provides a kind of Retinex Image Intensified System, and comprise iterative convolution module, reflecting component generation module, luminance component correcting module, reflecting component correcting module, strengthen Computer image genration module, above-mentioned module all realizes on FPGA;
Iterative convolution module output terminal is connected with the input end of reflecting component generation module with luminance component correcting module simultaneously, the output terminal of reflecting component generation module is connected with the input end of reflecting component correcting module, the output terminal of reflecting component correcting module is connected with the input end strengthening Computer image genration module, and the output terminal of luminance component correcting module is connected with the input end strengthening Computer image genration module.
The video source image feeding iterative convolution module of camera acquisition is carried out repeatedly mean filter grey iterative generation luminance component and is sent to reflecting component generation module and luminance component correcting module;
After reflecting component generation module receives luminance component, read computational reflect component after the source images being stored in first piece of random access memory, then reflecting component is sent to reflecting component correcting module;
After reflecting component correcting module receives reflecting component, utilize adaptive gray-level transform to revise reflecting component, revised reflecting component sends to and strengthens Computer image genration module;
After luminance component correcting module receives luminance component, utilize histogram equalization to revise luminance component, revised luminance component sends to and strengthens Computer image genration module;
Strengthen Computer image genration module and receive revised reflecting component and luminance component, generate the image after strengthening.
In Retinex Image Intensified System of the present invention, the middle partial data produced utilizes block random access memory in sheet to carry out buffer memory.Be specially: between the video source image of camera acquisition and reflecting component generation module, be connected with first piece of random access memory, reflecting component generation module calls first piece of random access memory, luminance component correcting module intrinsic call second piece of random access memory and the 3rd piece of random access memory, reflecting component generation module intrinsic call the 4th piece of random access memory and the 5th piece of random access memory.
The video source image of camera acquisition uses first piece of random access memory to carry out buffer memory; The gray-scale value number that luminance component correcting module carries out statistics with histogram is buffered in second piece of random access memory, and histogram equalization mapping table is buffered in the 3rd piece of random access memory; The gray-scale value number that reflecting component correcting module carries out self-adaptation gray scale stretching statistics is buffered in the 4th piece of random access memory, and self-adaptation gray scale stretching mapping table is buffered in the 5th piece of random access memory.
In iterative convolution module of the present invention, the mode of repeatedly mean filter iteration is adopted to approach the low pass convolution of Gaussian form around function G (x, y).As shown in Figure 2, each time after mean filter, judge whether to reach iterations, if do not reach iterations, this time mean filter result will be input to mean filter module, ask for mean filter value next time, iteration is gone down successively, until reach iterations.
The present invention is in iterative convolution module during calculation window, call fifo buffer FIFO, time delay is carried out to the source images F (x, y) of input, ensureing the timing synchronization of the view data of window interior, trying to achieve average by being averaged the view data summation of window interior.
The present invention is in iterative convolution module during calculation window, first the window (N is odd number) of N*N size is generated by fifo buffer FIFO, window is divided into simultaneously ((N-1)/2) * ((N-1)/2), (N-1) * 1, (N-1) * 1,1 new window of size, simultaneously to view data symmetrical fold summation in new window, finally by new window and summation, obtain whole window and; By divider computing, obtain mean filter value.Split window Folding sum, can reduce calculation step, reduces operation time, improves system real time.
In reflecting component generation module of the present invention, call the divider IP kernel that FPGA hardware design instrument ISE carries, this divider IP kernel, has stable performance, calculates accurately, adaptable feature, compared to other dividers, reduce system resource, the data precision is high, placement-and-routing's normative and reasonable, improves the stability of whole system.
In reflecting component correcting module of the present invention, statistics with histogram utilizes second piece of random access memory storage figure as reflecting component R (x, y) number of every one-level gray-scale value, according to the ratio that every one-level gray-scale value accounts in total gray level, gray-scale value corresponding after calculating histogram equalization, be stored in the 3rd piece of random access memory, be histogram equalization mapping table.
In luminance component correcting module of the present invention, self-adaptation gray scale stretches and utilizes the 4th piece of random access memory storage figure as luminance component I (x equally, y) number of every one-level gray-scale value, according to threshold range determination stretch ratio, then the ratio accounted in total gray level according to every one-level gray-scale value and stretch ratio, gray-scale value corresponding after calculating the stretching of self-adaptation gray scale, is stored in the 5th piece of random access memory, is self-adaptation gray scale stretching mapping table.
Embodiment
The hardware platform that the present embodiment realizes: the high-performance processor Spartan6LX150T chip adopting Xilinx company is in the processing system for video of core, adopt Verilog language to realize, programming and emulation complete on FPGA hardware design instrument ISE13.1.The video streaming image size of camera acquisition is 256*256 form.
Iterative convolution module
As Fig. 2, after iterative convolution module receives the video source image of camera acquisition, first the window (N is odd number) of N*N size is generated by fifo buffer FIFO, window is divided into ((N-1)/2) * ((N-1)/2), (N-1) * 1, (N-1) * 1,1 new window of size, simultaneously to view data symmetrical fold summation in new window, finally by new window and summation, obtain N*N size windows and; By divider computing, obtain first time mean filter value; Judge whether iterations reaches the threshold value of setting, if do not have, this mean filter value is sent into mean filter module again, until iterations reaches the threshold value of setting; Reflecting component generation module is sent on mean filter value (i.e. image illumination component) tunnel finally obtained, and luminance component correcting module is sent on a road.
In the present embodiment, Mean Filtering Algorithm adopts 9*9 window, and carry out 50 iteration altogether, the image illumination component of finally trying to achieve extremely approaches the low pass convolution results utilizing Gaussian form.The present invention changes window size and iterations according to the needs of realistic accuracy and the loss of resource.
Reflecting component generation module
Reflecting component connects after generation module receives image illumination component, by the source images in Clockreading first piece of random access memory, source image data position is converted to 16 bit data positions, call divider simultaneously, calculate the image reflecting component of 16, by arranging qualifications, what exceed gray level 255 gray-scale value is compressed to gray level 255, image reflecting component is limited to 8 bit data positions.
Luminance component correcting module
As Fig. 3, after system electrification, piece random access memory of second in statistics with histogram module is worth initialization, as the initial value of statistics with histogram with 0 by field sync signal.Subsequently when useful signal arrives, start to receive image illumination component (image intensity value), as Fig. 4, after statistics with histogram module receives this gray-scale value, two clock period are divided to complete statistics with histogram: first clock period, using receiving the address of next gray-scale value as histogram cache module, reads corresponding data; Second clock period writes histogram cache module respectively with same address after adding one to the data read.This module repeats said process to each effective gamma received, and can complete the statistics with histogram of each gray level of image illumination component.
After input control module has accepted a two field picture luminance component, produce an enable signal, histogram equalization mapping table generation module is driven to read the histogram Histogram module from 0 to 255 addresses successively, after order is cumulative, be multiplied with number of greyscale levels 255, and divided by total number of pixels, the gray-scale value after can being enhanced, and stored in piece random access memory of the 3rd in Well-Balanced Mapping table module, after reading 256 addresses, generate histogram equalization mapping table.In the implementation procedure of cumulative sum write, have employed water operation, greatly reducing the clock period number needed for whole process implementation.In this module, relate to multiplication and divide operations, utilize 255=256-1, and image source being 256*256, is 2
16, by displacement and subtraction, both saved logical resource on sheet, the arithmetic speed of FPGA can have been improved again, and then promoted the serviceability of whole system.
Reflecting component correcting module
In described image reflecting component module, after source images and image illumination component carry out computing, show that image reflecting component gives reflection correcting module, because the dynamic range of the image of reflecting component formation is less than normal, make output image can reach a good treatment effect, dynamic tensile will be carried out to reflecting component.First do statistics of histogram to reflecting component (image intensity value), under Low SNR, choosing compressibility factor is 5%, drops to minimum by the impact of blind element and noise.Minimum value in distribution search 10% maximum gradation value is as X
max, the maximal value in 15% minimum gradation value is as X
min.During stretching conversion, X will be greater than
maxpixel grey scale be set to Z
max, be less than X
minpixel grey scale be set to 0.Linear stretch interval is divided into [0, X by this algorithm adaptively
min), [X
min, X
max] and (X
max, 255] and three parts.Wherein [0, X
min) and (X
max, 255] and pixel grey scale between two gray areas is compressed to 0 and 255 respectively.If target is less in image, and target is just in time positioned at two by the interval compressed, and will be likely suppressed.For avoiding this situation to occur, the size of compressibility factor 5% optionally suitably can be adjusted.
When every two field picture reflecting component arrives, carry out statistics with histogram by the 4th piece of random access memory, record the number of times that each grey scale pixel value occurs, the X obtaining this two field picture can be added up
minand X
max.Because the adjacent two two field picture degrees of approximation are high, the X that available front frame obtains
minand X
maxprocess lower two field picture.Arrive to make at lower frame image data, [X (i, j)-X can be calculated
min] * 255/ [X
max-X
min], the IP kernel utilizing FPGA to carry completes, and is stored in the 5th piece of random access memory, realizes the real-time process to image.
Strengthen Computer image genration module
Read revised image illumination component and image reflecting component, according to clock, the image illumination component of corresponding pixel points is multiplied with image reflecting component, the image after being enhanced simultaneously, exports.
It is as shown in table 1 that the consumption of the present embodiment FPGA resource takies situation.As can be seen from Table 1, Retinex Image Intensified System of the present invention consumes sheet register resources on FPGA is 42%.Show that the present invention takies FPGA resource less, the resource space area of a room leaving other back-end algorithm for is larger; Resource consumption is rare is simultaneously beneficial to the inner cabling of FPGA, and the stability that the program that can ensure is run, improves robustness.
As can be seen from Figure 5, (a) source images for taking under insufficient light, smudgy, low visibility, brightness is on the low side.B () is (a) result through the process of successive ignition mean filter, detail section disappears substantially, only leaves luminance component.(c) reflecting component for producing after (a) filtering luminance component, but image is partially dark, and contrast strengthen and color recovery effects are also very limited, and integral image visual effect is not too natural.D (), for revising rear Retinex treatment effect, window railing is high-visible, and greenbelt more clearly can see the information such as branches and leaves, and electric pole and trunk can be told, automobile and background contrasts strengthen, and the task in the lower right corner also can be identified preferably, and overall visual experience improves.Thus, compared with prior art, the local detail of image strengthens to some extent in the present invention, further increases contrast and the sharpness of image, effectively overcomes the degeneration of light to image.
The consumption of table 1 embodiment hardware platform FPGA resource takies situation
| Resource classification | Use | Available | Utilization factor |
| Slice Registers | 4368 | 184308 | 2％ |
| Slice LUTs | 9918 | 92152 | 10％ |
| IOBs | 55 | 396 | 13％ |
| Block random access memory/FIFO | 113 | 268 | 42％ |
Claims (6)
1. a Retinex image enchancing method, is characterized in that, comprises the following steps:
Step one, mean filter grey iterative generation image illumination component is carried out to the source images of camera acquisition;
Step 2, obtain image reflecting component according to source images and image illumination component;
Step 3, use statistics with histogram method, according to the pixel number of each gray level of image illumination component, synthetic image irradiates the histogram of component; All gray-level pixels points are added up, accumulation result is divided equally in all gray levels, generate histogram equalization mapping table; By histogram equalization mapping table, generate the image illumination component after histogram equalization;
Step 4, use statistics with histogram method, according to the pixel number of each gray level of image reflecting component, the histogram of synthetic image reflecting component; According to pixel quantity threshold, read all gray-level pixels point and add up, calculate Adaptive Gray-Level scope and map gray level; By self organizing maps gray level, generate the image reflecting component after adaptive line conversion;
Step 5, according to revised image illumination component and image reflecting component, calculate and obtain image after Retinex image enhaucament.
2. one kind realizes the system of Retinex image enchancing method as claimed in claim 1, it is characterized in that, comprise iterative convolution module, reflecting component generation module, luminance component correcting module, reflecting component correcting module, strengthen Computer image genration module, above-mentioned module all realizes on FPGA;
Iterative convolution module output terminal is connected with the input end of reflecting component generation module with luminance component correcting module simultaneously, the output terminal of reflecting component generation module is connected with the input end of reflecting component correcting module, the output terminal of reflecting component correcting module is connected with the input end strengthening Computer image genration module, and the output terminal of luminance component correcting module is connected with the input end strengthening Computer image genration module;
The video streaming image feeding iterative convolution module of camera acquisition is carried out repeatedly mean filter grey iterative generation luminance component and is sent to reflecting component generation module and luminance component correcting module;
After reflecting component generation module receives luminance component, read computational reflect component after the source images being stored in first piece of random access memory, then reflecting component is sent to reflecting component correcting module;
After reflecting component correcting module receives reflecting component, utilize adaptive gray-level transform to revise reflecting component, revised reflecting component sends to and strengthens Computer image genration module;
After luminance component correcting module receives luminance component, utilize histogram equalization to revise luminance component, revised luminance component sends to and strengthens Computer image genration module;
Strengthen Computer image genration module and receive revised reflecting component and luminance component, generate the image after strengthening.
3. Retinex Image Intensified System as claimed in claim 2, it is characterized in that, first piece of random access memory is connected with between the video source image of camera acquisition and reflecting component generation module, reflecting component generation module calls first piece of random access memory, luminance component correcting module intrinsic call second piece of random access memory and the 3rd piece of random access memory, reflecting component generation module intrinsic call the 4th piece of random access memory and the 5th piece of random access memory; The video source image of camera acquisition uses first piece of random access memory to carry out buffer memory; The gray-scale value number that luminance component correcting module carries out statistics with histogram is buffered in second piece of random access memory, and histogram equalization mapping table is buffered in the 3rd piece of random access memory; The gray-scale value number that reflecting component correcting module carries out self-adaptation gray scale stretching statistics is buffered in the 4th piece of random access memory, and self-adaptation gray scale stretching mapping table is buffered in the 5th piece of random access memory.
4. Retinex Image Intensified System as claimed in claim 2, it is characterized in that, in iterative convolution module during calculation window, first the window of N*N size is generated by fifo buffer FIFO, N is odd number, window is divided into simultaneously ((N-1)/2) * ((N-1)/2), (N-1) * 1, (N-1) * 1,1 new window of size; Then to view data symmetrical fold summation in new window, finally by new window and summation, obtain whole window and; By divider computing, obtain mean filter value.
5. Retinex Image Intensified System as claimed in claim 2, it is characterized in that, in reflecting component correcting module, statistics with histogram utilizes second piece of random access memory storage figure as the number of the every one-level gray-scale value of reflecting component, according to the ratio that every one-level gray-scale value accounts in total gray level, gray-scale value corresponding after calculating histogram equalization, is stored in the 3rd piece of random access memory, generates histogram equalization mapping table.
6. Retinex Image Intensified System as claimed in claim 2, it is characterized in that, in luminance component correcting module, self-adaptation gray scale stretches and utilizes the 4th piece of random access memory storage figure as the number of the every one-level gray-scale value of luminance component, according to threshold range determination stretch ratio, then the ratio accounted in total gray level according to every one-level gray-scale value and stretch ratio, gray-scale value corresponding after calculating the stretching of self-adaptation gray scale, be stored in the 5th piece of random access memory, generate self-adaptation gray scale stretching mapping table.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410450328.3A CN105469373B (en) | 2014-09-04 | 2014-09-04 | A kind of Retinex image enchancing methods and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410450328.3A CN105469373B (en) | 2014-09-04 | 2014-09-04 | A kind of Retinex image enchancing methods and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105469373A true CN105469373A (en) | 2016-04-06 |
| CN105469373B CN105469373B (en) | 2018-07-31 |
Family
ID=55607034
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410450328.3A Active CN105469373B (en) | 2014-09-04 | 2014-09-04 | A kind of Retinex image enchancing methods and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105469373B (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106169081A (en) * | 2016-06-29 | 2016-11-30 | 北京工业大学 | A kind of image classification based on different illumination and processing method |
| CN106327446A (en) * | 2016-08-25 | 2017-01-11 | 中国科学院长春光学精密机械与物理研究所 | Image processing method and device |
| CN106780375A (en) * | 2016-12-02 | 2017-05-31 | 南京邮电大学 | A kind of image enchancing method under low-light (level) environment |
| CN106780379A (en) * | 2016-12-08 | 2017-05-31 | 哈尔滨工业大学 | The microscopical colour-image reinforcing method of one kind metering |
| CN106845359A (en) * | 2016-12-26 | 2017-06-13 | 四川农业大学 | Tunnel portal driving prompt apparatus and method based on infrared emission |
| CN107330871A (en) * | 2017-06-29 | 2017-11-07 | 西安工程大学 | The image enchancing method of insulator automatic identification is run under bad weather condition |
| CN108595379A (en) * | 2018-05-08 | 2018-09-28 | 济南浪潮高新科技投资发展有限公司 | A kind of parallelization convolution algorithm method and system based on multi-level buffer |
| CN108648151A (en) * | 2018-05-09 | 2018-10-12 | 电子科技大学 | A kind of license plate horizontally tilted automatic correction method under bad weather |
| CN109300101A (en) * | 2018-10-18 | 2019-02-01 | 重庆邮电大学 | A kind of more exposure image fusion methods based on Retinex theory |
| CN109345586A (en) * | 2018-11-02 | 2019-02-15 | 国网湖南省电力有限公司 | Electrical equipment discharge characteristic extracting method based on ultraviolet imagery technology |
| CN109584143A (en) * | 2018-12-07 | 2019-04-05 | 中国科学院长春光学精密机械与物理研究所 | A kind of aerial camera image-enhancing equipment and method |
| CN111107274A (en) * | 2018-10-26 | 2020-05-05 | 北京图森智途科技有限公司 | Image brightness statistical method and imaging device |
| CN111260588A (en) * | 2020-02-13 | 2020-06-09 | 苏州盖德光电科技有限公司 | Image enhancement method for high-definition digital CMOS imaging assembly |
| WO2020186372A1 (en) * | 2019-03-18 | 2020-09-24 | 北京大学深圳研究生院 | Inverse tone mapping method, system, and device, and computer readable medium |
| CN112837250A (en) * | 2021-01-27 | 2021-05-25 | 武汉华中数控股份有限公司 | Infrared image self-adaptive enhancement method based on generalized histogram equalization |
| WO2021120790A1 (en) * | 2019-12-19 | 2021-06-24 | 腾讯科技（深圳）有限公司 | Image enhancement method and apparatus, and computer-readable storage medium |
| CN113781360A (en) * | 2021-09-27 | 2021-12-10 | 朱明辉 | Infusion device image enhancement method and system based on artificial intelligence |
| CN116028013A (en) * | 2023-02-20 | 2023-04-28 | 辰星（天津）自动化设备有限公司 | Optimization method, device, divider and medium for FPGA division operation |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120106867A1 (en) * | 2010-10-29 | 2012-05-03 | JVC Kenwood Corporation | Image processing apparatus and image processing method |
| CN103413275A (en) * | 2013-07-26 | 2013-11-27 | 北京工业大学 | Retinex night image enhancement method based on gradient zero norm minimum |
-
2014
- 2014-09-04 CN CN201410450328.3A patent/CN105469373B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120106867A1 (en) * | 2010-10-29 | 2012-05-03 | JVC Kenwood Corporation | Image processing apparatus and image processing method |
| CN103413275A (en) * | 2013-07-26 | 2013-11-27 | 北京工业大学 | Retinex night image enhancement method based on gradient zero norm minimum |
Non-Patent Citations (3)
| Title |
|---|
| 王科俊 等: "基于窗口无关均值滤波的MSR图像增强", 《哈尔滨工业大学学报》 * |
| 詹筱: "高动态范围红外图像压缩的细节增强算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 陈雾: "基于Retinex理论的图像增强算法研究", 《中国优秀博硕士学位论文全文数据库 (硕士) 信息科技辑》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106169081B (en) * | 2016-06-29 | 2019-07-05 | 北京工业大学 | A kind of image classification and processing method based on different illumination |
| CN106169081A (en) * | 2016-06-29 | 2016-11-30 | 北京工业大学 | A kind of image classification based on different illumination and processing method |
| CN106327446A (en) * | 2016-08-25 | 2017-01-11 | 中国科学院长春光学精密机械与物理研究所 | Image processing method and device |
| CN106327446B (en) * | 2016-08-25 | 2017-10-27 | 中国科学院长春光学精密机械与物理研究所 | A kind of image processing method and device |
| CN106780375A (en) * | 2016-12-02 | 2017-05-31 | 南京邮电大学 | A kind of image enchancing method under low-light (level) environment |
| CN106780379A (en) * | 2016-12-08 | 2017-05-31 | 哈尔滨工业大学 | The microscopical colour-image reinforcing method of one kind metering |
| CN106780379B (en) * | 2016-12-08 | 2019-08-23 | 哈尔滨工业大学 | A kind of microscopical colour-image reinforcing method of metering |
| CN106845359A (en) * | 2016-12-26 | 2017-06-13 | 四川农业大学 | Tunnel portal driving prompt apparatus and method based on infrared emission |
| CN107330871A (en) * | 2017-06-29 | 2017-11-07 | 西安工程大学 | The image enchancing method of insulator automatic identification is run under bad weather condition |
| CN108595379A (en) * | 2018-05-08 | 2018-09-28 | 济南浪潮高新科技投资发展有限公司 | A kind of parallelization convolution algorithm method and system based on multi-level buffer |
| CN108648151A (en) * | 2018-05-09 | 2018-10-12 | 电子科技大学 | A kind of license plate horizontally tilted automatic correction method under bad weather |
| CN109300101A (en) * | 2018-10-18 | 2019-02-01 | 重庆邮电大学 | A kind of more exposure image fusion methods based on Retinex theory |
| CN111107274A (en) * | 2018-10-26 | 2020-05-05 | 北京图森智途科技有限公司 | Image brightness statistical method and imaging device |
| CN111107274B (en) * | 2018-10-26 | 2021-01-08 | 北京图森智途科技有限公司 | Image brightness statistical method and imaging device |
| US11875486B2 (en) | 2018-10-26 | 2024-01-16 | Beijing Tusen Zhitu Technology Co., Ltd. | Image brightness statistical method and imaging device |
| CN109345586A (en) * | 2018-11-02 | 2019-02-15 | 国网湖南省电力有限公司 | Electrical equipment discharge characteristic extracting method based on ultraviolet imagery technology |
| CN109584143A (en) * | 2018-12-07 | 2019-04-05 | 中国科学院长春光学精密机械与物理研究所 | A kind of aerial camera image-enhancing equipment and method |
| WO2020186372A1 (en) * | 2019-03-18 | 2020-09-24 | 北京大学深圳研究生院 | Inverse tone mapping method, system, and device, and computer readable medium |
| WO2021120790A1 (en) * | 2019-12-19 | 2021-06-24 | 腾讯科技（深圳）有限公司 | Image enhancement method and apparatus, and computer-readable storage medium |
| CN111260588A (en) * | 2020-02-13 | 2020-06-09 | 苏州盖德光电科技有限公司 | Image enhancement method for high-definition digital CMOS imaging assembly |
| CN112837250A (en) * | 2021-01-27 | 2021-05-25 | 武汉华中数控股份有限公司 | Infrared image self-adaptive enhancement method based on generalized histogram equalization |
| CN113781360A (en) * | 2021-09-27 | 2021-12-10 | 朱明辉 | Infusion device image enhancement method and system based on artificial intelligence |
| CN113781360B (en) * | 2021-09-27 | 2023-10-20 | 牟宗勇 | Image enhancement method and system for infusion device based on artificial intelligence |
| CN116028013A (en) * | 2023-02-20 | 2023-04-28 | 辰星（天津）自动化设备有限公司 | Optimization method, device, divider and medium for FPGA division operation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105469373B (en) | 2018-07-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105469373A (en) | Retinex image enhancing method and system | |
| Gao et al. | Naturalness preserved nonuniform illumination estimation for image enhancement based on retinex | |
| US11467661B2 (en) | Gaze-point determining method, contrast adjusting method, and contrast adjusting apparatus, virtual reality device and storage medium | |
| CN103295191B (en) | Multiple scale vision method for adaptive image enhancement and evaluation method | |
| CN104268843B (en) | Image self-adapting enhancement method based on histogram modification | |
| CN103530847A (en) | Infrared image enhancing method | |
| CN107358586A (en) | A kind of image enchancing method, device and equipment | |
| Lai et al. | Improved local histogram equalization with gradient-based weighting process for edge preservation | |
| CN108805829B (en) | Image data processing method, device, equipment and computer readable storage medium | |
| CN103501401B (en) | Towards super large noise real-time video based on pre-filtering denoising method | |
| CN103985097B (en) | A kind of image enchancing method of Nonuniform illumination | |
| CN105046658A (en) | Low-illumination image processing method and device | |
| CN106169181A (en) | A kind of image processing method and system | |
| Khan et al. | Tone-mapping using perceptual-quantizer and image histogram | |
| CN109919859A (en) | A kind of Outdoor Scene image defogging Enhancement Method calculates equipment and its storage medium | |
| CN103295205B (en) | A kind of low-light-level image quick enhancement method based on Retinex and device | |
| CN103839245B (en) | The Retinex colour-image reinforcing method at night of Corpus--based Method rule | |
| Tan et al. | A real-time video denoising algorithm with FPGA implementation for Poisson–Gaussian noise | |
| CN110349112A (en) | A kind of two stages image de-noising method based on adaptive singular value threshold value | |
| CN104794689A (en) | Preprocessing method for enhancing sonar image contract | |
| CN111242872B (en) | Real-time RGB image enhancement method based on MSR | |
| CN117197627B (en) | Multi-mode image fusion method based on high-order degradation model | |
| CN103996179A (en) | Fast real-time image enhancement method based on single-scale Retinex | |
| CN114070959A (en) | FPGA-based video denoising hardware implementation method | |
| CN116645291A (en) | Self-adaptive Gamma correction glare suppression method based on regional brightness perception |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |