CN111311503A - Night low-brightness image enhancement system - Google Patents
Night low-brightness image enhancement system Download PDFInfo
- Publication number
- CN111311503A CN111311503A CN201911106602.4A CN201911106602A CN111311503A CN 111311503 A CN111311503 A CN 111311503A CN 201911106602 A CN201911106602 A CN 201911106602A CN 111311503 A CN111311503 A CN 111311503A
- Authority
- CN
- China
- Prior art keywords
- image
- brightness
- image enhancement
- unit
- night
- 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.)
- Pending
Links
- 238000013507 mapping Methods 0.000 claims abstract description 22
- 238000005457 optimization Methods 0.000 claims abstract description 13
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000006243 chemical reaction Methods 0.000 claims abstract description 9
- 230000009467 reduction Effects 0.000 claims abstract description 8
- 238000001914 filtration Methods 0.000 claims description 33
- 238000001514 detection method Methods 0.000 claims description 23
- 230000002776 aggregation Effects 0.000 claims description 8
- 238000004220 aggregation Methods 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000010845 search algorithm Methods 0.000 claims description 3
- 230000026676 system process Effects 0.000 claims description 3
- 239000003623 enhancer Substances 0.000 claims description 2
- 238000000034 method Methods 0.000 abstract description 29
- 230000000694 effects Effects 0.000 abstract description 12
- 230000009466 transformation Effects 0.000 description 8
- 238000005286 illumination Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 235000008733 Citrus aurantifolia Nutrition 0.000 description 3
- 235000011941 Tilia x europaea Nutrition 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 239000004571 lime Substances 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 230000004438 eyesight Effects 0.000 description 2
- 238000003707 image sharpening Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000006116 polymerization reaction Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000001035 drying Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000004297 night vision Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G06T5/70—
-
- G06T5/73—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Abstract
The application belongs to the technical field of image processing, and particularly relates to a night low-brightness image enhancement system. The existing night low-brightness image enhancement method is directly applied to an unmanned aerial vehicle cruise system, and has the problems of insufficient real-time performance, insufficient robustness, generation of extra noise, smoothness of images and the like. The application provides a night low-brightness image enhancement system, which comprises an image enhancement subsystem and an image denoising and sharpening subsystem, wherein the image enhancement subsystem comprises an image mapping and stretching unit and an input image scale factor value optimization unit, and the image denoising and sharpening subsystem comprises a color space conversion unit and a three-dimensional block matching unit. The brightness enhancement of different scenes with different brightness is realized, the noise reduction and sharpening effect is realized, and the robustness of the algorithm is highest.
Description
Technical Field
The application belongs to the technical field of image processing, and particularly relates to a night low-brightness image enhancement system.
Background
Night city public security management is an important link of smart city construction, in recent years, the technical development of unmanned aerial vehicles is rapid, and the application of unmanned aerial vehicles to night city public security management is the trend of future city management. The image quality is a key influencing factor for urban public security management at night, and because the night illumination is insufficient, no matter professional cameras or consumer-grade cameras can not shoot satisfactory night images, the night image enhancement technology becomes a research hotspot of scholars at home and abroad in recent years. In the night picture of the unmanned aerial vehicle, the readability of the image directly influences the subsequent identification and detection.
Brightness enhancement is an important method for improving image quality, and is widely applied to the fields of underwater image reconstruction, medical image enhancement and intelligent transportation. However, the difference between the application field of the system and the night unmanned aerial vehicle cruise system is large, for example, the difference between underwater image enhancement and the actual scene on land is large, and the medical field application mainly aims at the research that objects are small and the precision reaches the micro-nano level; the application in the fields of unmanned driving and intelligent traffic is fixed on road traffic. The above are directed to their major professional areas, with high dependence on lighting and environmental conditions. However, in the actual unmanned aerial vehicle cruise intelligent detection system, the related scenes are many, such as factory forbidden land, traffic tracking, night theft prevention, dark corner pedestrian discrimination and other aspects, the interference of the external environment easily causes unstable illumination, the brightness fluctuation is large, the noise condition is not clear, the difficulty of city public security management is increased, and especially the remote condition of the monitoring place from the unmanned aerial vehicle is increased. For complex application scenes, some night image enhancement algorithms with good adaptability are proposed. Shen J et al propose a retinex method by adaptive attenuation quantization to enhance brightness detail (AAQR); gupta B et al propose enhancing images with gamma correction and weighted cumulative probability distribution functions for pixels (AGCCPF); guo X et al propose a defogging transformation map model to optimize the initial illumination map and thus better reconstruct the illumination map (LIME); FuX et al propose adjusting image brightness (MF) by fusing multiple order derivatives of the initial estimate of the illumination map; jiang B et al propose separating a reflectance map and an illumination Map (MSRCR) using multi-scale gaussian filtering with color reconstruction; fu X et al use the estimated reflectance and illumination components to change the brightness.
The existing night low-brightness image enhancement method is directly applied to an unmanned aerial vehicle cruise system, and has the problems of insufficient real-time performance, insufficient robustness, generation of extra noise, smoothness of images and the like.
Disclosure of Invention
1. Technical problem to be solved
Based on the fact that the existing night low-brightness image enhancement method is directly applied to an unmanned aerial vehicle cruise system, the problems that real-time performance is insufficient, robustness is insufficient, extra noise and image smoothness can be generated and the like exist.
2. Technical scheme
In order to achieve the above-mentioned object, the present application provides a night low brightness image enhancement system, comprising an image enhancement subsystem and an image noise reduction sharpening subsystem,
the image enhancer system includes an image mapping stretching unit and a scale factor value optimizing unit of an input image,
the image mapping and stretching unit is used for carrying out enhancement processing on the image,
a scale factor value optimization unit for adjusting the sharpness of the image according to the brightness of the image,
the image denoising and sharpening subsystem comprises a color space conversion unit and a three-dimensional block matching unit,
the color space conversion unit is used for converting the RGB color space of the image into the YCbCr color space,
and the three-dimensional block matching unit is used for realizing the noise reduction and sharpening of the image.
The present application provides another embodiment as follows: the image mapping and stretching unit is a hyperbolic tangent curve mapping and stretching unit, and when the domain of definition of the hyperbolic tangent curve is larger than zero, high brightness is restrained, and low brightness is stretched.
The present application provides another embodiment as follows: and the hyperbolic tangent curve mapping and stretching unit is used for weighting the image enhancement intensity according to the intensity of the input image and lightening the enhancement processing of a high-brightness area.
The present application provides another embodiment as follows: and the hyperbolic tangent curve mapping and stretching unit stretches the weighted image in a dynamic range.
The present application provides another embodiment as follows: the weighting coefficient is the average value of R, G and B channels of the image pixel points.
The present application provides another embodiment as follows: the scale factor value optimization unit of the input image is based on a golden section search algorithm, and the scale factor value range of the input image is [0.1, 9.9 ].
The present application provides another embodiment as follows: the three-dimensional block matching unit comprises a basic estimation module, a final estimation module and an unsharp filtering module.
The present application provides another embodiment as follows: the basic estimation module includes a first block matching sub-module, a collaborative hard threshold filtering sub-module, and a first aggregation sub-module.
The present application provides another embodiment as follows: the final estimation module comprises a second block matching sub-module, a collaborative Wiener filtering sub-module and a second aggregation sub-module.
The present application provides another embodiment as follows: after the image enhancement system processes the image, the image F value, the image confidence coefficient, the image patch-based contrast quality index and the image contrast are all improved, wherein the image F value comprehensively reflects the accuracy rate and the recall rate of target detection and is a comprehensive evaluation index of the target detection performance.
3. Advantageous effects
Compared with the prior art, the night low-brightness image enhancement system provided by the application has the beneficial effects that:
the application provides a night low-brightness image enhancement system, to unmanned aerial vehicle vision intelligent detection system night, study advanced night low-brightness image enhancement algorithm, provided a night low-brightness image enhancement method towards unmanned aerial vehicle vision intelligent detection, improved unmanned aerial vehicle cruise monitoring system's image quality night, and then improved night target identification and detection rate of accuracy, be favorable to promoting the high-efficient management of local peace.
The application provides a night low-brightness image enhancement system, provides a night image enhancement method integrating an optimal hyperbolic tangent curve and a modified three-dimensional matching block for urban public security management at night of an unmanned aerial vehicle, and is more suitable for an unmanned aerial vehicle night cruise system.
The night low-brightness image enhancement system provided by the application is oriented to an intelligent night unmanned aerial vehicle visual detection system, and provides a night image enhancement method combining an optimal hyperbolic tangent curve and an improved three-dimensional matching block; the optimization of parameters is realized by adopting a golden section algorithm to maximize the entropy of the image; secondly, an enhancement method for improving three-dimensional block matching by combining an unsharp filtering technology in a YCbCr space is provided, and the image denoising and sharpening processing is realized.
Compared with the current advanced night low-brightness image enhancement method, the night low-brightness image enhancement system provided by the application has the advantages that the target identification accuracy and the algorithm running time are in the leading position in the unmanned aerial vehicle intelligent detection system. The robustness of different types of images is better represented, and the real-time performance can meet the requirements of practical application.
The application provides a low-luminance image enhancement system night turns into YCrCb color space from RGB color space to go on processing space, directly carries out luminance to the Y passageway and handles, has avoided going dry the influence of sharpening process to the color component.
According to the night low-brightness image enhancement system, the problem of image smoothness in a three-dimensional block matching algorithm is solved through unsharp mask filtering, and dryness removal and sharpening of a night low-brightness image are achieved.
The night low-brightness image enhancement system comprehensively considers image brightness and image target recognition effects, and performs brightness enhancement on low-brightness images through the optimal hyperbolic tangent curve. The three-dimensional block matching algorithm is improved through unsharp mask filtering, so that dryness removal and sharpening of the image are realized, and the accuracy of target identification is improved.
The night low-brightness image enhancement system provided by the application adopts the golden section algorithm to perform self-adaptive optimization in the k value optimization process, and divides the k values into three groups through image brightness selection so as to improve the operation efficiency of the algorithm.
Drawings
FIG. 1 is a schematic diagram of a night low brightness image enhancement system of the present application;
FIG. 2 is a schematic diagram of the non-sharpening filtering based BM3D algorithm of the present application;
FIG. 3 is a diagram of pedestrian recognition effects at night before and after enhancement;
FIG. 4 is a diagram illustrating the results of various enhancement methods for pedestrian identification;
in the figure: the image processing method comprises the steps of 1-an image enhancement subsystem, 2-an image denoising and sharpening subsystem, 3-an image mapping and stretching unit, 4-an optimization unit of a scale factor value of an input image, 5-a color space conversion unit, 6-a three-dimensional block matching unit, 7-a basic estimation module, 8-a final estimation module, 9-an unsharp filtering module, 10-a first block matching submodule, 11-a collaborative hard threshold filtering submodule, 12-a first aggregation submodule, 13-a second block matching submodule, 14-a collaborative wiener filtering submodule and 15-a second aggregation submodule.
Detailed Description
Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, and it will be apparent to those skilled in the art from this detailed description that the present application can be practiced. Features from different embodiments may be combined to yield new embodiments, or certain features may be substituted for certain embodiments to yield yet further preferred embodiments, without departing from the principles of the present application.
AAQR can have obvious over-enhancement, AGCCPF pays attention to maintaining average brightness, black noise occurs, LIME can have smooth image, enhancement effect in some hidden areas is poor, MF and MSRCR have unobvious advantages in real-time indexes, and some scenes can generate red noise. SRIE runs too long to achieve real-time performance.
Referring to fig. 1 to 4, the present application provides a night low-brightness image enhancement system, which includes an image enhancement subsystem 1 and an image denoising and sharpening subsystem 2,
the image enhancement subsystem 1 comprises an image mapping and stretching unit 3 and an input image scale factor value optimization unit 4, the image mapping and stretching unit 3 is used for enhancing the image,
a scale factor value optimization unit 4 for the input image for adjusting the sharpness of the image according to the brightness of the image,
the image denoising and sharpening subsystem 2 comprises a color space conversion unit 5 and a three-dimensional block matching unit 6,
the color space conversion unit 5 is used for converting the image RGB color space into YCbCr color space,
and the three-dimensional block matching unit 6 is used for realizing the dryness reduction and sharpening of the image.
Further, the image mapping and stretching unit 3 is a hyperbolic tangent curve mapping and stretching unit, and when the domain of definition of the hyperbolic tangent curve is greater than zero, the high brightness is suppressed, and the low brightness is stretched.
Further, the hyperbolic tangent curve mapping and stretching unit weights the image enhancement intensity according to the intensity of the input image, and reduces the enhancement processing of the high-luminance region.
Further, the hyperbolic tangent curve mapping and stretching unit stretches the weighted image in a dynamic range.
Further, the weighting coefficients are R, G and B three-channel average values of image pixel points.
When the hyperbolic tangent curve is larger than zero in the definition domain, the hyperbolic tangent curve has the characteristics of high brightness inhibition and low brightness stretching, can be effectively applied to low-brightness image enhancement at night, and avoids excessive image enhancement. The parameters of the low-brightness image enhancement method based on the optimal hyperbolic tangent function can be adjusted in a self-adaptive mode, the optimal parameters of the golden section algorithm can be used, artifacts are not generated, and the method has general adaptability. (1) Hyperbolic tangent curve mapping stretching image
Let the input color image collected by the drone be i (x) ═ { r (x), g (x), b (x) }, R, G, B are the three color components of the image, respectively, and they are normalized to [0, 1 ]. Where x represents the position of an image pixel, including the horizontal and vertical coordinates x, y. Then, nonlinear mapping is performed by using the property of the hyperbolic tangent function, and the image brightness is enhanced. The expression of the hyperbolic tangent function is:
where phi is i (x) the scaling of the image intensity, i.e., phi is k × i (x), where k is the scale factor of the input image.
To avoid over-enhancement, the image enhancement intensity should be weighted according to the intensity of the input image to mitigate the enhancement processing in the high brightness region. The weighted output expression is:
Iw(x)＝w×I(x)+(1-wD×tanh(k×I(x)) (2)
where w is a weighting coefficient of the input image. In order to keep the reality of the image to the maximum extent, the three-channel average value of the image pixel points is used as a weight (w).
In order to realize enhancement of a global image by widely distributing the enhancement effect to the entire gray scale, the weighted image is stretched in the dynamic range. So that the stretched image is
Wherein, Iw(x)maxIs a weighted image Iw(x) Maximum gray value of Iw(x)minIs a weighted image Iw(x) Is measured. Combining equations (2) and (3), the final enhancement formula is obtained as follows:
as can be seen from the equation (4), the final unknown parameter is only k, and a proper k value is crucial to the image enhancement result.
Further, the scale factor value optimization unit of the input image is based on a golden section search algorithm, and the scale factor value range of the input image is [0.1, 9.9 ].
The entropy of the image is a characterization parameter of the average information quantity of the image, and the larger the entropy value is, the more the information quantity carried by the image is, and the clearer the image is. The paper adopts a golden section algorithm to gradually approach the maximum image entropy, and the corresponding k value is the optimal value at the moment. The definition of the image entropy is:
where p (i) is the probability that the intensity i appears in the image.
The method of selecting the initial range of k values according to the brightness of the image is adopted, and the brightness is defined as follows:
wherein
X, Y are the number of image rows and columns, and X represents the pixel coordinate.
In an unmanned aerial vehicle night image enhancement system, according to the brightness of an image, the operation efficiency of an algorithm and the actual test condition are considered, and the following method defined k value initial range is determined:
the three groups of conditions respectively represent the night image with almost no light, the night image with light in a small area and the night image with normal light.
Further, the three-dimensional block matching unit comprises a basic estimation module, a final estimation module and an unsharp filtering module.
Further, the basic estimation module includes a first block matching sub-module, a collaborative hard threshold filtering sub-module, and a first aggregation sub-module.
Further, the final estimation module includes a second block matching sub-module, a collaborative Wiener filtering sub-module, and a second aggregation sub-module.
BM3D removes the dryness is based on the image denoising strategy of transform domain enhancement sparse representation, can some details in the better reservation image, through searching similar block and filtering in the transform domain, obtains the block evaluation value, finally carries on the weighting to every point in the image and obtains the final denoising effect. However, after the drying is removed, image blurring exists in different degrees, for this reason, the non-sharpening filtering algorithm is researched and applied to the three-dimensional block matching algorithm, and in order to avoid influencing the color of the image while further improving the image quality at night, the image in the RGB color space is converted into the YCbCr color space for processing.
The YCbCr color space is a color space representation in which Y represents the luminance component, Cb represents the blue chrominance component, and Cr represents the red chrominance component.
(1) Color space conversion
In the process of carrying out noise reduction and sharpening on the image, in order to improve the operating efficiency and avoid influencing color components, the RGB color space of the image is converted into the YCbCr color space, Y represents the brightness information of the image, Cb and Cr represent the color information of the image, and the subsequent enhancement processing is carried out on a Y channel, so that the influence of the image on the color information is avoided, the number of channels is reduced, and the efficiency of an algorithm can also be improved. The formula for converting the RGB color space to the YCbCr color space is as follows:
(2) three-dimensional block matching algorithm based on unsharp filtering
The non-sharpening filtering can realize image sharpening, the non-sharpening filtering can be applied to BM3D to effectively avoid image blurring in the denoising process, the BM3D algorithm based on the non-sharpening filtering is mainly divided into three steps, namely basic estimation, final estimation and non-sharpening filtering, and the specific flow is shown in FIG. 2
A) Basic estimation
1) Block matching: and searching similar blocks of the target area, and superposing the two-dimensional image blocks into a three-dimensional array.
2) Collaborative hard threshold filtering: and carrying out three-dimensional transformation on the matched three-dimensional image blocks to obtain sparse representation of the image group, reducing noise by carrying out hard threshold filtering on the transformed sparse representation, then carrying out three-dimensional inverse transformation to generate estimation of all the grouping blocks, and returning to an image domain. It should be noted that the term "cooperate" is a literal meaning, and means that each grouping segment cooperatively filters all other segments, because a three-dimensional segment is formed by combining two-dimensional segments through block matching, then three-dimensional transformation is performed, and then filtering operation is performed, and the whole process is that other segments are cooperatively filtered among the segments.
3) Polymerization: multiple estimated values may appear in the same similar block after three-dimensional transformation, and a weighted average method is needed to obtain a final estimated value.
B) And (3) final estimation:
improved grouping and collaborative Wiener filtering is performed using the base estimate as input.
1) Block matching: the location of similar blocks is found in the basic estimation using a block matching approach. Using these positions, two three-dimensional clusters can be obtained, one from the noisy image and one from the basic estimated image.
2) Collaborative Wiener filtering: and performing three-dimensional transformation on the two groups to obtain transformation coefficients, taking the basically estimated three-dimensional groups as energy spectrums of real signals, and performing collaborative wiener filtering on the noise images by using the energy spectrums. And performing three-dimensional inverse transformation on the filter coefficients, deducing final estimated values of all the grouping blocks, and returning to the original positions of the similar blocks.
3) Polymerization: and summarizing all obtained local estimation values by using a weighted average method, and calculating a final estimation value of the real image.
C) Unsharp filtering:
after BM3D final estimation is performed on the Y channel, sharpening is performed on the image by unsharp filtering, and the image is converted back to the RGB color space to obtain a final enhanced image. The unsharp mask filtering is mainly used for obtaining an unsharp image of the image, subtracting unsharp low-frequency information from the image difference to further obtain a high-frequency component of the image, namely a correction signal, and then overlapping the correction signal with an original image to obtain a final enhanced image.
Further, after the image enhancement system processes the image, the image F value, the image confidence coefficient, the image patch-based contrast quality index and the image contrast are all improved, wherein the image F value comprehensively reflects the accuracy rate and the recall rate of target detection and is a comprehensive evaluation index of the target detection performance.
Most night low-brightness image enhancement algorithms are applied to narrow professional fields, such as the underwater field, the medical field and the traffic field, and are not applicable to complex environments where an unmanned aerial vehicle intelligent detection system needs to adapt to multiple scenes, night low-brightness unmanned aerial vehicle image enhancement can be realized in the application, pedestrian recognition is carried out on images before and after enhancement by adopting a YOLOv3 model, and the result shows that the F value after image enhancement is improved by 46%, the confidence coefficient is improved by 38%, the PCQI (contrast quality index based on patches) is improved by 0.35, and the contrast is improved by 0.02.
Compared with the prior advanced method, the night low-brightness image enhancement system provided by the application has better running time performance of the algorithm, the six comparison algorithms and the text algorithm are respectively adopted for 160 images, and the average running time and the number of frames per second of each image are shown in the table 1. It can be seen that the running time of AGCCPF is the shortest, the running time of each graph is 0.14 second, 7.1 frames are run per second, the running times of AAQR, LIME and the method herein are close, the number of frames per second is 3-4 frames, while the running times of the other three methods are longer, the real-time performance is poor, the running time of SRIE reaches 10.93 seconds, and the real-time enhancement is difficult to achieve. Therefore, this application can reach unmanned aerial vehicle intelligent detection system's real-time requirement, the operating time preferred. From the enhancement effect, all advanced night enhancement methods can improve the brightness of night images, but some methods have over-enhancement and some methods have black noise points with poor visual quality after enhancement; some enhancement is followed by more severe red noise. Compared with the prior art, the system can better realize the brightness enhancement of different scenes with different brightness, simultaneously realize the effect of noise reduction and sharpening, and has the highest robustness of the algorithm.
TABLE 1 run times of different algorithms
Selection of experimental data: three groups of pictures of normal light, darker light and extremely dark light are selected according to the brightness of the light; two groups of pictures of a road and a dark road are selected according to a scene area, a single target and a crowd dense area are selected according to the target sparsity, in addition, the condition of a small target at a distance is considered, 8 groups of images are obtained totally, 20 images are collected in each group, 160 images are collected in total, and the size of an input image is 480x 360. An experimental platform of all data in the application is matlab2018a, a computer system is 64-bit windows7, a processor is InterCore i5-6500 and 3.2GHz, an integrated display card is used, an operating memory is 8G, the average operating time of the algorithm is 0.26 second, and each second is 3.85 frames.
Pedestrian target recognition is carried out by adopting a YOLOv3 model, so that the enhancement effect of the application is verified. Fig. 3 is a pedestrian detection result of an image before and after enhancement by using the YOLOv3 model, and it can be seen that after the system of the present application is enhanced, the image brightness is greatly improved, and the target information is clearer. The effectiveness of the application is also verified from the view of the pedestrian identification detection result. In the target identification detection, the F value is often used for objectively evaluating the identification detection effect, the different groups of unmanned aerial vehicle images in the table 2 display the pedestrian detection F value and the confidence table, after the application, the enhancement effect of low-brightness images at various nights is obvious, and the pedestrian detection F value and the confidence degree are obviously improved.
Fig. 3 shows the recognition result of the original image, fig. 4 shows the recognition result after the processing by the text method, and the numbers beside the gray boxes indicate the confidence.
Table 2 different groups of unmanned aerial vehicle image pedestrian detection F-value and confidence table
This application can not only be suitable for unmanned aerial vehicle night vision detection system, can be applicable to the enhancement of the night low-brightness image under the complex environment moreover, and the comprehensive consideration luminance enhancement and image sharpening remove dryness. For other scene areas, brightness enhancement can be realized only by adjusting the k value optimization parameter according to the scene brightness. In addition, the hyperbolic tangent curve can realize low-brightness stretching and high-brightness suppression in a part of a defined domain larger than zero, and by utilizing the property, the image brightness can be enhanced by using other corresponding functions such as a region of a sine function [0, pi/2 ].
Although the present application has been described above with reference to specific embodiments, those skilled in the art will recognize that many changes may be made in the configuration and details of the present application within the principles and scope of the present application. The scope of protection of the application is determined by the appended claims, and all changes that come within the meaning and range of equivalency of the technical features are intended to be embraced therein.
Claims (10)
1. A night low brightness image enhancement system, comprising: comprises an image enhancement subsystem and an image noise reduction and sharpening subsystem,
the image enhancer system includes an image mapping stretching unit and a scale factor value optimizing unit of an input image,
the image mapping and stretching unit is used for carrying out enhancement processing on the image,
a scale factor value optimization unit for adjusting the sharpness of the image according to the brightness of the image,
the image denoising and sharpening subsystem comprises a color space conversion unit and a three-dimensional block matching unit,
the color space conversion unit is used for converting the RGB color space of the image into the YCbCr color space,
and the three-dimensional block matching unit is used for realizing the noise reduction and sharpening of the image.
2. The night low brightness image enhancement system of claim 1 wherein: the image mapping and stretching unit is a hyperbolic tangent curve mapping and stretching unit, and when the domain of definition of the hyperbolic tangent curve is larger than zero, high brightness is restrained, and low brightness is stretched.
3. The night low brightness image enhancement system of claim 2 wherein: and the hyperbolic tangent curve mapping and stretching unit is used for weighting the image enhancement intensity according to the intensity of the input image and lightening the enhancement processing of a high-brightness area.
4. The night low brightness image enhancement system of claim 3 wherein: and the hyperbolic tangent curve mapping and stretching unit stretches the weighted image in a dynamic range.
5. The night low brightness image enhancement system of claim 3 wherein: the weighting coefficient is the average value of R, G and B channels of the image pixel points.
6. The night low brightness image enhancement system of claim 1 wherein: the scale factor value optimization unit of the input image is based on a golden section search algorithm, and the scale factor value range of the input image is [0.1, 9.9 ].
7. The night low brightness image enhancement system of claim 1 wherein: the three-dimensional block matching unit comprises a basic estimation module, a final estimation module and an unsharp filtering module.
8. The night low brightness image enhancement system of claim 7 wherein: the basic estimation module includes a first block matching sub-module, a collaborative hard threshold filtering sub-module, and a first aggregation sub-module.
9. The night low brightness image enhancement system of claim 7 wherein: the final estimation module comprises a second block matching sub-module, a collaborative wiener filtering sub-module and a second aggregation sub-module.
10. The night low brightness image enhancement system of any one of claims 1 to 9 wherein: after the image enhancement system processes the image, the image F value, the image confidence coefficient, the contrast quality index of the image based on the patch and the image contrast are all improved;
the image F value comprehensively reflects the accuracy rate and the recall rate of target detection and is a comprehensive evaluation index of the target detection performance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911106602.4A CN111311503A (en) | 2019-11-13 | 2019-11-13 | Night low-brightness image enhancement system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911106602.4A CN111311503A (en) | 2019-11-13 | 2019-11-13 | Night low-brightness image enhancement system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111311503A true CN111311503A (en) | 2020-06-19 |
Family
ID=71146749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911106602.4A Pending CN111311503A (en) | 2019-11-13 | 2019-11-13 | Night low-brightness image enhancement system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111311503A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113393389A (en) * | 2021-06-02 | 2021-09-14 | 武汉博宇光电系统有限责任公司 | Image enhancement method without artificial halation |
| CN114283156A (en) * | 2021-12-02 | 2022-04-05 | 珠海移科智能科技有限公司 | Method and device for removing document image color and handwriting |
| CN114445300A (en) * | 2022-01-29 | 2022-05-06 | 赵恒� | Nonlinear underwater image gain algorithm for hyperbolic tangent deformation function transformation |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102640184A (en) * | 2009-10-20 | 2012-08-15 | 苹果公司 | Temporal filtering techniques for image signal processing |
| US20130071025A1 (en) * | 2011-09-21 | 2013-03-21 | Industry-University Cooperation Foundation Sogang University | Image processing method and image processing apparatus |
| CN105931208A (en) * | 2015-08-22 | 2016-09-07 | 首都师范大学 | Physical model-based low-illuminance image enhancement algorithm |
-
2019
- 2019-11-13 CN CN201911106602.4A patent/CN111311503A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102640184A (en) * | 2009-10-20 | 2012-08-15 | 苹果公司 | Temporal filtering techniques for image signal processing |
| US20130071025A1 (en) * | 2011-09-21 | 2013-03-21 | Industry-University Cooperation Foundation Sogang University | Image processing method and image processing apparatus |
| CN105931208A (en) * | 2015-08-22 | 2016-09-07 | 首都师范大学 | Physical model-based low-illuminance image enhancement algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 戴丽: "《中国优秀硕士学位论文全文数据库》", 15 March 2015 * |
| 王伟江 等: "面向机柜表面缺陷检测的不均匀光照和低亮度图像增强方法", 《仪器仪表学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113393389A (en) * | 2021-06-02 | 2021-09-14 | 武汉博宇光电系统有限责任公司 | Image enhancement method without artificial halation |
| CN114283156A (en) * | 2021-12-02 | 2022-04-05 | 珠海移科智能科技有限公司 | Method and device for removing document image color and handwriting |
| CN114283156B (en) * | 2021-12-02 | 2024-03-05 | 珠海移科智能科技有限公司 | Method and device for removing document image color and handwriting |
| CN114445300A (en) * | 2022-01-29 | 2022-05-06 | 赵恒� | Nonlinear underwater image gain algorithm for hyperbolic tangent deformation function transformation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Fast haze removal for nighttime image using maximum reflectance prior | |
| US10339643B2 (en) | Algorithm and device for image processing | |
| Huang et al. | An efficient visibility enhancement algorithm for road scenes captured by intelligent transportation systems | |
| Park et al. | Single image dehazing with image entropy and information fidelity | |
| CN103914813B (en) | The restored method of colored haze image defogging and illumination compensation | |
| CN107527332A (en) | Enhancement Method is kept based on the low-light (level) image color for improving Retinex | |
| Gao et al. | Sand-dust image restoration based on reversing the blue channel prior | |
| Ma et al. | An effective fusion defogging approach for single sea fog image | |
| CN102665034A (en) | Night effect removal method for camera-collected video | |
| CN111311503A (en) | Night low-brightness image enhancement system | |
| Wang et al. | Variational single nighttime image haze removal with a gray haze-line prior | |
| CN106651817A (en) | Non-sampling contourlet-based image enhancement method | |
| CN112116536A (en) | Low-illumination image enhancement method and system | |
| Li et al. | Low illumination video image enhancement | |
| CN106846258A (en) | A kind of single image to the fog method based on weighted least squares filtering | |
| Parihar et al. | A comprehensive analysis of fusion-based image enhancement techniques | |
| Lin et al. | Adaptive infrared and visible image fusion method by using rolling guidance filter and saliency detection | |
| Wei et al. | An image fusion dehazing algorithm based on dark channel prior and retinex | |
| CN109345479B (en) | Real-time preprocessing method and storage medium for video monitoring data | |
| Honda et al. | Make my day-high-fidelity color denoising with near-infrared | |
| CN116823686B (en) | Night infrared and visible light image fusion method based on image enhancement | |
| Hmue et al. | Image enhancement and quality assessment methods in turbid water: A review article | |
| CN107644403B (en) | Method for correcting non-uniform color of image under severe environment condition | |
| CN116263942A (en) | Method for adjusting image contrast, storage medium and computer program product | |
| CN112822343B (en) | Night video oriented sharpening method and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |