CN114494063A - Night traffic image enhancement method based on biological vision mechanism - Google Patents

Night traffic image enhancement method based on biological vision mechanism Download PDF

Info

Publication number
CN114494063A
CN114494063A CN202210087269.2A CN202210087269A CN114494063A CN 114494063 A CN114494063 A CN 114494063A CN 202210087269 A CN202210087269 A CN 202210087269A CN 114494063 A CN114494063 A CN 114494063A
Authority
CN
China
Prior art keywords
image
frequency
brightness
night traffic
enhanced
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
Application number
CN202210087269.2A
Other languages
Chinese (zh)
Other versions
CN114494063B (en
Inventor
颜红梅
何亚辉
张显石
吴章碧
任伟
孙仟禧
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202210087269.2A priority Critical patent/CN114494063B/en
Publication of CN114494063A publication Critical patent/CN114494063A/en
Application granted granted Critical
Publication of CN114494063B publication Critical patent/CN114494063B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G06T5/94—Dynamic range modification of images or parts thereof based on local image properties, e.g. for local contrast enhancement
    • 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
    • G06T2207/10048—Infrared image
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a night traffic image enhancement method based on a biological vision mechanism, which is applied to the technical field of computer vision and aims at the image problems of dim night road environment, interference of opposite running vehicle light on the sight of a driver, fuzziness of pedestrians and street lamps and the like, so that the night traffic image brightness is improved, and night driving tasks are easier and safer; the method comprises the steps of firstly constructing a night traffic image classifier, dividing the night traffic image into three classes according to the brightness and the local contrast of the night traffic image, then calling three different image enhancement algorithms according to different classification results to enhance the input night traffic image, and finally evaluating the quality of the image before and after enhancement without reference image.

Description

Night traffic image enhancement method based on biological vision mechanism
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to a night traffic image enhancement technology.
Background
With the development of the assistant driving technology, when the automatic driving technology is still immature and people drive at night and go out frequently, people desire the advanced assistant driving technology to improve the comfort and safety of driving at night. In the related field of assistant driving technology, night traffic image enhancement is a hot topic, and for a darker driving environment, a vehicle-mounted optical imaging sensing system cannot accurately sense surrounding environment information, so that an assistant driving system has the possibility of making wrong decisions, potential safety hazards are brought to night driving, and therefore the night traffic image enhancement plays a vital role in night assistant driving. In fact, the night traffic environment is a dynamic scene, the lighting conditions in different scenes are different, and it is not only an examination for drivers but also a challenge for advanced driving assistance systems to accurately capture information in driving scenes. The method is characterized in that the method is inspired by an early biological vision dual-channel, the cone and rod cells on the retina transmit nerve impulses to horizontal cells after receiving light stimulation, the horizontal cells transmit the nerve impulses to bipolar cells, the activated degrees of the horizontal cells and the bipolar cells are different under the influence of the intensity of the light stimulation in the process of layer-by-layer transmission, and therefore different image enhancement algorithms are selected to enhance the dynamically-changed traffic scene images.
As is well known, compared with driving under the conditions of daytime and good weather conditions, driving at night has dim surrounding environment, the interference of street lamps and lamps of vehicles driving to each other and pedestrians in the sight range caused by dimming can increase the danger of driving at night, the interference of light makes drivers more prone to visual fatigue, and the dimming of environment makes drivers focus more attention. Therefore, after the current driving situation and the demand at night are fully known, it is necessary to design an image enhancement method which can be changed along with the change of traffic scenes, so that the driving assistance system can obtain images with the best visual quality to assist the driver in driving no matter what night scene the driver drives, and the safety and the comfort of driving at night are improved.
Disclosure of Invention
In order to solve the technical problem, the invention provides a night traffic image enhancement method based on a biological vision mechanism.
The technical scheme adopted by the invention is as follows: a night traffic image enhancement method based on a biological vision mechanism comprises the following steps:
s1, acquiring night traffic images of different scenes by using a vehicle event data recorder, constructing a data set, and dividing the data set according to the brightness of the images;
s2, according to the classification of the images in the S1 and the brightness and local contrast of the images, taking the images as vectors to train a Support Vector Machine (SVM) based on a Gaussian radial basis kernel function so as to construct a night traffic image classifier;
and S3, classifying the images according to the night traffic image classifier established in the step S2, calling a corresponding image enhancement algorithm for enhancing the night traffic images in each category, and finally outputting the enhanced images.
The dividing of the data set according to the brightness of the image in step S1 specifically includes:
dividing the image with the brightness value larger than 185400 into: category "01";
dividing an image having a luminance value greater than 121500 and less than or equal to 185400 into: the "02" category;
dividing the image with the brightness value larger than 0 and smaller than 121500 into: the "03" category.
The method further comprises the following steps of calculating the brightness value of the image, wherein the calculation process comprises the following steps: and converting the image from the RGB channel to the HSV channel, extracting a V channel image, and summing all pixels of the V channel image to obtain the brightness value of the image.
Step S2 specifically includes: the classified class information of the image is used as a label, the brightness and the local contrast of the image are used as the characteristics of the image, and a support vector machine based on a Gaussian radial basis kernel function is trained by using the vector formed by the characteristics of the image and the label, so that the night traffic image classifier is obtained.
The implementation process of the image enhancement algorithm of the class 01 image is as follows: carrying out image decomposition on an image to be enhanced to obtain a high-frequency detail component and a low-frequency brightness component;
the high-frequency detail component is processed by adopting a noise suppression strategy;
the processing procedure of the low-frequency brightness component is as follows: firstly, converting an image from an RGB channel to an HSV channel, and then carrying out logarithmic transformation on the V channel image; then converting the HSV channel image back into an RGB channel image;
and finally, overlapping the processed high-frequency detail component and the low-frequency brightness component to obtain an enhanced image.
The noise suppression strategy is realized by the following steps: convolving the high-frequency detail components decomposed from the image with a two-dimensional Gaussian function to obtain a weight map, and multiplying the weight map and the decomposed high-frequency detail components pixel by pixel to obtain the high-frequency image after noise suppression.
The implementation process of the image enhancement algorithm of the 02-class image is as follows: carrying out image decomposition on an image to be enhanced to obtain a high-frequency detail component and a low-frequency brightness component;
the high-frequency detail component is processed by adopting a noise suppression strategy;
the processing procedure of the low-frequency brightness component is as follows: RGB is adopted for sampling, the image sampling of three channels simulates rod cells and cone cells in biological visual processing,
Figure BDA0003488405140000021
Figure BDA0003488405140000022
wherein f isR(x，y)、fG(x，y)、fB(x, y) are the image samples of the three channels, HC, respectivelyinSignals representing input level cells, HCbackIs an output image after horizontal cell processing,
Figure BDA0003488405140000031
horizontal cells, σ, modeled for a two-dimensional Gaussian functionL(x，y)、σc(x, y) is the variance of the Gaussian function;
enhanced control of cone signals by horizontal cells is described by the Naka-Rushton empirical formula:
Figure BDA0003488405140000032
Figure BDA0003488405140000033
Figure BDA0003488405140000034
as output of horizontal cells, i.e. HCback；
The horizontal cells then transmit the enhanced signal to the bipolar cells, which output an enhanced low frequency image BCout containing luminancec(x，y)：
Figure BDA0003488405140000035
Figure BDA0003488405140000036
Wherein ". sup." denotes a convolution,
Figure BDA0003488405140000037
for a bipolar cell central receptive field modeled with a two-dimensional gaussian function,
Figure BDA0003488405140000038
for a bipolar peripheral receptive field, k, modeled using a two-dimensional Gaussian functionIs a value range of [0, 1]The weight of (2);
and finally, overlapping the processed high-frequency detail component and the low-frequency brightness component to obtain an enhanced image.
σL(x，y)、σcThe value of (x, y) is determined in the following manner:
Figure BDA0003488405140000039
where n is L or c, m is the average luminance of all pixels, s is the standard deviation of all pixels, sigma is a predefined parameter,
Figure BDA00034884051400000310
representing pixel points in the processed image.
The implementation process of the image enhancement algorithm of the '03' class image is as follows: carrying out image decomposition on an image to be enhanced to obtain a high-frequency detail component and a low-frequency brightness component;
the high-frequency detail component is processed by adopting a noise suppression strategy;
the processing procedure of the low-frequency brightness component is as follows: channel conversion is carried out on the RGB image to obtain an HSV image, and a V channel image is extracted to carry out self-adaptive gamma correction; the self-adaptive gamma correction realizes the gamma correction of the image by taking a gamma value in a self-adaptive way according to the brightness of the image, and the gamma value is as follows:
Figure BDA0003488405140000041
wherein
Figure BDA0003488405140000042
Wherein, IinFor inputting the luminance channel image to be enhanced, IoutThe luminance channel image is corrected and enhanced, and lum is the luminance value of the image;
then, after the V channel is enhanced, the HSV image is restored into an RGB image;
and finally, overlapping the processed high-frequency detail component and the low-frequency brightness component to obtain an enhanced image.
The invention has the beneficial effects that: the method classifies night traffic images based on the traditional machine learning classification algorithm and the biological vision mechanism, and then calls the image enhancement algorithm of the corresponding category according to the category of the images to enhance the night traffic images, so that clear scenes of night traffic scenes are obtained, drivers are helped to effectively find possible dangers in advance, the driving burden is reduced, and the driving safety and the driving comfort are improved; the method can be used in combination with an advanced assistant driving system, and the decision level of the advanced assistant driving system is improved.
Drawings
Fig. 1 is a flow chart of the nighttime traffic image classification and enhancement task of the present invention.
FIG. 2 is a flow chart of the night traffic image classifier training and testing of the present invention.
Fig. 3 is a diagram of the classification of an image data set into three categories according to image brightness.
FIG. 4 is a flow chart of an algorithm for enhancing class 01 nighttime traffic images in accordance with the present invention;
wherein (a) is a process flow diagram of the present invention; (b) a logarithmic function used for enhancing the brightness of the low-frequency channel image after image decomposition; (c) the values of the numerical transformation curve are respectively taken from different v values.
FIG. 5 is a flow chart of the enhancement algorithm for the "02" class night traffic image of the present invention.
Fig. 6 is a flowchart of an algorithm for enhancing a "03" class night traffic image according to the present invention.
FIG. 7 is a graph of the result of enhancing a night time traffic image using the method of the present invention;
the following description is made of the drawings, in which (a) is a first original image, (b) is a second original image, (c) is a third original image, (d) is an image enhanced by the method of the present invention, (e) is an image enhanced by the method of the present invention, (f) is an image enhanced by the method of the present invention, (g) is a local image in the first original image, (h) is a local image in the second original image, (i) is a local image in the third original image, (j) is an effect of enhancing a local image in the first original image by the method of the present invention, (k) is an effect of enhancing a local image in the second original image by the method of the present invention, and (l) is an effect of enhancing a local image in the third original image by the method of the present invention.
FIG. 8 is a statistical chart of the quantitative index results before and after 200 nighttime traffic images are enhanced by the method of the present invention;
wherein, (a) is the comprehensive image quality evaluation function of the image of the '01' type, (b) is the comprehensive image quality evaluation function of the image of the '02' type, (c) is the comprehensive image quality evaluation function of the image of the '03' type, (d) is the IL-NIQE of the image of the '01' type, (e) is the IL-NIQE of the image of the '02' type, (f) is the IL-NIQE of the image of the '03' type, (g) is the quality index based on the spatial spectral entropy of the image of the '01' type, (h) is the quality index based on the spatial spectral entropy of the image of the '02' type, and (i) is the quality index based on the spatial spectral entropy of the image of the '03' type.
Detailed Description
In order to facilitate the understanding of the technical contents of the present invention by those skilled in the art, the present invention will be further explained with reference to the accompanying drawings.
The implementation process of the invention comprises the following four steps:
A. and (3) carrying out image category division according to the brightness and the local contrast of the night traffic image: the method includes the steps of calculating brightness and local contrast of an image in a data set by using a data set made of acquired data, reflecting the image data set on a two-dimensional coordinate by using the brightness of the image as an abscissa and the local contrast as an ordinate, and manually setting two brightness thresholds to divide the whole data set into three categories, for example, in the embodiment, two thresholds 121500 and 185400 are set according to brightness values, and then dividing the categories into the following specific categories:
the image class whose luminance value lum is greater than 185400 is the "01" class;
the image category whose luminance value is greater than 121500 and less than or equal to 185400 is the "02" category;
the image class whose luminance value is greater than 0 and less than 121500 is the "03" class.
The brightness value lum is calculated by calculating the sum of all pixel values of the V-channel image after the image is converted from the RGB space to the HSV space. Wherein, the brightness of the category 01 is higher, the brightness of the category 02 is moderate, and the brightness of the category 03 is lowest. The division results are shown in fig. 3.
Step A1, the calculation process of the brightness value of the night traffic image is as follows: and converting the RGB image into an HSV channel image, extracting the V channel image, and summing all pixels of the V channel image to obtain the brightness value of the image.
Step A2, the calculation process of the local contrast value of the night traffic image is as follows: converting the RGB image into an HSV channel image, extracting the V channel image, calculating the standard deviation of pixels in a window of 21 x 21 of the V channel image to obtain a local contrast map of the image, and then calculating the sum of all pixels of the local contrast map to obtain the local contrast value of the image.
Step A3, setting two threshold values related to the brightness value of the image, and dividing the brightness value into three categories;
the 01 category image is characterized in that bright and dazzling light sources are more, light fields are complex, and local contrast is good (namely image details are clear) in an image scene; the '02' type image is characterized in that the brightness of each part in the image is moderate, no special dazzling light source exists, even if a part of the image has an area which cannot be seen clearly, the part has a pixel value, the local contrast is good, and the details are clear; the '03' type image is characterized in that the overall brightness of the image is very dark, the pixel values of a large number of pixels in the image are 0, and the local contrast is poor.
B. Construction of night traffic image classifier
FIG. 2 illustrates a night traffic image classifier training and testing process. The classified class information of the image classification is used as a label, the brightness and the local contrast map of the image are used as the characteristics of the image, and a Support Vector Machine (SVM) based on a Gaussian radial basis kernel function is trained by using the characteristics and the vector formed by the label, so that the night traffic image classifier is obtained.
C. Night traffic image enhancement algorithm:
the invention adopts the overall thought that: and adopting a classification enhancement strategy for the night traffic image. Fig. 4, 5 and 6 are flow charts of algorithms for enhancing three types of images, respectively. Fig. 4 shows the image enhancement algorithm for the "01" class image as: carrying out image decomposition on an input image to be enhanced; the invention adopts an image decomposition method based on total-variation energy (total-variation energy), which is proposed in 2006 by J.F.Aujol et al; decomposing an input image into a high-frequency detail component containing image details and noise and a low-frequency brightness component containing image brightness, wherein a noise suppression strategy is adopted in the high-frequency part, and is realized as follows: convolving the high-frequency image decomposed by the image with a two-dimensional Gaussian function to obtain a weight map, and then multiplying the weight map and the decomposed high-frequency image pixel by pixel to obtain a noise-suppressed high-frequency path output component, wherein the expression is as follows:
Figure BDA0003488405140000061
Figure BDA0003488405140000062
Figure BDA0003488405140000063
wherein, muC(x, y) is a high-frequency image
Figure BDA0003488405140000064
A weight image obtained by convolution with a two-dimensional Gaussian function G (x, y),
Figure BDA0003488405140000065
is a high frequency path image of the noise suppressed output.
The low-frequency brightness component firstly converts the image from an RGB channel to an HSV channel, and then carries out logarithmic transformation on the V channel image so as to inhibit a highlight area and promote a low-brightness area; and then converting the HSV channel image into an RGB image, and finally superposing the images processed by the high-frequency path and the low-frequency path to obtain an enhanced image. The logarithmic transformation expression employed in fig. 4 is:
S＝logv+1(1+v*r)
s is an image output after logarithmic transformation, r represents a pixel value of the image, v is a parameter value for controlling image enhancement intensity, and the smaller the v value is, the larger the image luminance enhancement is, but we find through experiments that the smaller the v value is, the better the image luminance enhancement is, and the relationship between the image luminance enhancement and the v value is as shown in fig. 4(c), and we find that the visual effect of image enhancement is the best when v is 30 according to a large number of image enhancement experiments.
Fig. 5 shows the image enhancement algorithm for the "02" class image as follows: decomposing an input enhanced image into high frequency detail components IdetailAnd a low-frequency luminance component IbaseThe processing of the high-frequency detail component adopts a noise suppression strategy, and the high-frequency image after noise suppression is recorded as Outdetail(ii) a The low-frequency brightness component is sampled by RGB, the image sampling of three channels simulates rod cells and cone cells in biological visual processing,
Figure BDA0003488405140000071
Figure BDA0003488405140000072
wherein f isR(x，y)，fG(x, y) and fB(x, y) are the light signal sampling of the cone and rod, respectively, reflected to the image processing, i.e. RGB sampling of the image, so the signal of the input horizontal cell HC is HCin，HCbackIs an output image after horizontal cell processing,
Figure BDA0003488405140000073
horizontal cells, σ, modeled for a two-dimensional Gaussian functionL(x，y)、σc(x, y) is the variance of the Gaussian function;
horizontal cell receptive field size is regulated by the level of brightness:
Figure BDA0003488405140000074
m is the average luminance of all pixels, s is the standard deviation of all pixels, sigma is the predefined parameter set to 1,
Figure BDA0003488405140000075
representing pixel points in the processed image; enhanced control of cone signals by horizontal cells is described by the Naka-Rushton empirical formula:
Figure BDA0003488405140000081
Figure BDA0003488405140000082
Figure BDA0003488405140000083
as output of horizontal cells, i.e. HCback；
The horizontal cells then transmit the enhanced signal to bipolar cells BC whose receptive field consists of a smaller central excitatory domain and a larger peripheral inhibitory domain, a structure generally described as a gaussian difference model. The bipolar cell utilizes the antagonism mechanism of the center and the periphery, can effectively reduce the redundancy of input signals, and increases the spatial resolution and the contrast sensitivity
Figure BDA0003488405140000084
Figure BDA0003488405140000085
Here, the，BCoutc(x, y) represents the low frequency image containing the luminance after image enhancement, "+" represents the convolution,
Figure BDA0003488405140000086
for a bipolar cell central receptive field modeled with a two-dimensional gaussian function,
Figure BDA0003488405140000087
for a bipolar cell peripheral receptive field modeled by a two-dimensional Gaussian function, k is a value range of [0, 1 ]]The relative sizes of the center and periphery of the cell are double-hit. The established night traffic image enhancement model of the biological vision mechanism enables the enhanced image to be more in line with a human visual sense system; out in FIG. 5baseRepresenting the low-frequency image containing brightness after image enhancement;
and finally, overlapping the high-frequency path and the low-frequency path to obtain an enhanced image.
Fig. 6 shows the image enhancement algorithm for the "03" class image as follows: decomposing high-frequency details and low-frequency brightness components of an image to be enhanced, wherein the high-frequency details still adopt a noise suppression strategy for processing, and the low-frequency brightness still adopts a noise suppression strategy for channel conversion of an RGB image into an HSV image, and extracting a V channel image for self-adaptive gamma correction; the adaptive gamma correction is realized by adaptively taking a gamma value according to the brightness (lun) of an image to realize the gamma correction of the image, wherein 4 gamma values which are set by us are as follows:
Figure BDA0003488405140000088
wherein
Figure BDA0003488405140000089
Wherein, IinFor inputting the luminance channel image to be enhanced, IoutTo correct the enhanced luminance channel image, lum is the sum of all pixels of the luminance channel (V-channel) image (i.e., the luminance value of the image).
And finally, after the V channel is enhanced, the HSV image is restored into an RGB image, and the high-frequency component and the low-frequency component are superposed to obtain the enhanced image.
D. Testing and evaluating the effects of the night traffic image classifier and the night traffic image enhancement algorithm:
step D1: 200 pieces of image data of the night traffic image classifier are respectively placed in three folders named by categories "01", "02" and "03". The accuracy of the obtained test result is 92%;
step D2: fig. 7 shows the enhanced visual effect of the three types of night traffic images, which are provided by the invention, the brightness and the local contrast of the three types of images are obviously improved compared with the original image, and objects originally blurred in the dark in the image scene become clear and visible, so that the visual quality of the images is improved, the driving of a driver can be effectively assisted, and the safety of driving at night is improved. FIG. 8 is an evaluation of image quality before and after image enhancement by using three night traffic image non-reference image evaluation indexes, wherein the adopted quantitative indexes are a comprehensive image quality evaluation function (vpmi), IL-NIQE and a quality index (SSEQ) based on spatial spectral entropy, wherein the larger the vpmi value is, the better the image quality is, and the smaller the IL-NIQE and SSEQ values are, the better the image quality is. The abscissa n in the figure indicates the number of images; fig. 8 also proves the effectiveness of classifying and enhancing the night traffic image by the method of the present invention from quantitative indexes, and it can be seen from the figure that the value of the comprehensive image quality evaluation function (vpmi) of the enhanced image is obviously higher than that of the original image, the value of IL-NIQE is obviously smaller than that of the original image, and the value of the quality index (SSEQ) based on the spatial spectral entropy is obviously smaller than that of the original image.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims (9)

1. A night traffic image enhancement method based on a biological vision mechanism is characterized by comprising the following steps:
s1, acquiring night traffic images of different scenes by using a vehicle event data recorder, constructing a data set, and dividing the data set according to the brightness of the images;
s2, according to the classification of the images in the S1 and the brightness and local contrast of the images, taking the images as vectors to train a Support Vector Machine (SVM) based on a Gaussian radial basis kernel function so as to construct a night traffic image classifier;
and S3, classifying the images according to the night traffic image classifier established in the step S2, calling a corresponding image enhancement algorithm for enhancing the night traffic images in each category, and finally outputting the enhanced images.
2. The night traffic image enhancement method based on biological visual mechanism according to claim 1, wherein the step S1 is to divide the data set according to the brightness of the image, specifically:
dividing the image with the brightness value larger than 185400 into: category "01";
dividing an image having a luminance value greater than 121500 and less than or equal to 185400 into: the "02" category;
dividing the image with the brightness value larger than 0 and smaller than 121500 into: the "03" category.
3. The night traffic image enhancement method based on the biological vision mechanism as claimed in claim 2, further comprising calculating brightness values of the images by: and converting the image from the RGB channel to the HSV channel, extracting a V channel image, and summing all pixels of the V channel image to obtain the brightness value of the image.
4. The night traffic image enhancement method based on biological visual mechanism according to claim 3, wherein the step S2 is specifically as follows: the classified class information of the image is used as a label, the brightness and the local contrast of the image are used as the characteristics of the image, and a support vector machine based on a Gaussian radial basis kernel function is trained by using the vector formed by the characteristics of the image and the label, so that the night traffic image classifier is obtained.
5. The night traffic image enhancement method based on biological vision mechanism as claimed in claim 4, wherein the image enhancement algorithm of the "01" class image is implemented as follows: carrying out image decomposition on an image to be enhanced to obtain a high-frequency detail component and a low-frequency brightness component;
the high-frequency detail component is processed by adopting a noise suppression strategy;
the processing procedure of the low-frequency brightness component is as follows: firstly, converting an image from an RGB channel to an HSV channel, and then carrying out logarithmic transformation on the V channel image; then converting the HSV channel image back into an RGB channel image;
and finally, overlapping the processed high-frequency detail component and the low-frequency brightness component to obtain an enhanced image.
6. The night traffic image enhancement method based on the biological vision mechanism as claimed in claim 4, wherein the image enhancement algorithm of the "02" class image is implemented as follows: carrying out image decomposition on an image to be enhanced to obtain a high-frequency detail component and a low-frequency brightness component;
the high-frequency detail component is processed by adopting a noise suppression strategy;
the processing procedure of the low-frequency brightness component is as follows: RGB is adopted for sampling, the image sampling of three channels simulates rod cells and cone cells in biological visual processing,
Figure FDA0003488405130000021
Figure FDA0003488405130000022
wherein f isR(x，y)、fG(x，y)、fB(x, y) are the image samples of the three channels, HC, respectivelyinSignals representing input level cells, HCbackIs an output image after horizontal cell processing,
Figure FDA0003488405130000023
horizontal cells, σ, modeled for a two-dimensional Gaussian functionL(x，y)、σc(x, y) is the variance of the Gaussian function;
enhanced control of cone signals by horizontal cells is described by the Naka-Rushton empirical formula:
Figure FDA0003488405130000024
Figure FDA0003488405130000025
Figure FDA0003488405130000026
as output of horizontal cells, i.e. HCback；
The horizontal cells then transmit the enhanced signal to the bipolar cells, which output an enhanced low frequency image BCout containing luminancec(x，y)：
Figure FDA0003488405130000027
Figure FDA0003488405130000028
Wherein ". sup." denotes a convolution,
Figure FDA0003488405130000029
for a bipolar cell central receptive field modeled with a two-dimensional gaussian function,
Figure FDA00034884051300000210
for a bipolar cell peripheral receptive field modeled by a two-dimensional Gaussian function, k is a value range of [0, 1 ]]The weight of (2);
and finally, overlapping the processed high-frequency detail component and the low-frequency brightness component to obtain an enhanced image.
7. The night traffic image enhancement method based on biological vision mechanism according to claim 6, characterized in that σ isL(x，y)、σcThe value of (x, y) is determined in the following manner:
Figure FDA0003488405130000031
where n is L or c, m is the average luminance of all pixels, s is the standard deviation of all pixels, sigma is a predefined parameter,
Figure FDA0003488405130000032
representing pixel points in the processed image.
8. The night traffic image enhancement method based on the biological vision mechanism as claimed in claim 4, wherein the image enhancement algorithm of the "03" class image is implemented as follows: carrying out image decomposition on an image to be enhanced to obtain a high-frequency detail component and a low-frequency brightness component;
the high-frequency detail component is processed by adopting a noise suppression strategy;
the processing procedure of the low-frequency brightness component is as follows: channel conversion is carried out on the RGB image to obtain an HSV image, and a V channel image is extracted to carry out self-adaptive gamma correction; the self-adaptive gamma correction realizes the gamma correction of the image by taking a gamma value in a self-adaptive way according to the brightness of the image, and the gamma value is as follows:
Figure FDA0003488405130000033
wherein
Figure FDA0003488405130000034
Wherein, IinFor inputting the luminance channel image to be enhanced, IoutThe luminance channel image is corrected and enhanced, and lum is the luminance value of the image;
then, after the V channel is enhanced, the HSV image is restored into an RGB image;
and finally, overlapping the processed high-frequency detail component and the low-frequency brightness component to obtain an enhanced image.
9. The night traffic image enhancement method based on biological visual mechanism according to claim 5, 6 or 8, characterized in that the noise suppression strategy is implemented as follows: convolving the high-frequency detail components decomposed from the image with a two-dimensional Gaussian function to obtain a weight map, and multiplying the weight map and the decomposed high-frequency detail components pixel by pixel to obtain the high-frequency image after noise suppression.
CN202210087269.2A 2022-01-25 2022-01-25 Night traffic image enhancement method based on biological vision mechanism Active CN114494063B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210087269.2A CN114494063B (en) 2022-01-25 2022-01-25 Night traffic image enhancement method based on biological vision mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210087269.2A CN114494063B (en) 2022-01-25 2022-01-25 Night traffic image enhancement method based on biological vision mechanism

Publications (2)

Publication Number Publication Date
CN114494063A true CN114494063A (en) 2022-05-13
CN114494063B CN114494063B (en) 2023-04-07

Family

ID=81475039

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210087269.2A Active CN114494063B (en) 2022-01-25 2022-01-25 Night traffic image enhancement method based on biological vision mechanism

Country Status (1)

Country Link
CN (1) CN114494063B (en)

Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110181787A1 (en) * 2010-01-28 2011-07-28 Stmicroelectronics Asia Pacific Pte Ltd Image details preservation and enhancement
CN103065159A (en) * 2013-01-14 2013-04-24 合肥工业大学 Image classification method based on brightness and contrast ratio
US20170193643A1 (en) * 2014-09-30 2017-07-06 Fujifilm Corporation Image processing apparatus, image processing method, program, and recording medium
US20180227509A1 (en) * 2015-08-05 2018-08-09 Wuhan Guide Infrared Co., Ltd. Visible light image and infrared image fusion processing system and fusion method
CN109691980A (en) * 2019-01-07 2019-04-30 哈尔滨理工大学 A kind of diabetic retina image lesion detection method
CN109978787A (en) * 2019-03-25 2019-07-05 电子科技大学 A kind of image processing method based on biological vision computation model
US20200082367A1 (en) * 2018-09-07 2020-03-12 Bank Of America Corporation Processing System for Providing a Teller Assistant Experience Using Enhanced Reality Interfaces
CN111161181A (en) * 2019-12-26 2020-05-15 深圳市优必选科技股份有限公司 Image data enhancement method, model training method, device and storage medium
CN111652817A (en) * 2020-05-28 2020-09-11 大连海事大学 Underwater image sharpening method based on human eye visual perception mechanism
CN111652806A (en) * 2019-03-15 2020-09-11 上海铼锶信息技术有限公司 Method and system for removing shadow of image
CN112287861A (en) * 2020-11-05 2021-01-29 山东交通学院 Road information enhancement and driving early warning method based on night environment perception
CN112481839A (en) * 2020-11-27 2021-03-12 杭州数迈科技有限公司 Fabric sewing method, system, equipment and storage medium
WO2021111152A1 (en) * 2019-12-06 2021-06-10 Ibex Innovations Ltd Composition-guided post processing for x-ray images
CN113139929A (en) * 2021-03-17 2021-07-20 浙江工业大学 Gastrointestinal tract endoscope image preprocessing method comprising information screening and fusion repairing
CN113674159A (en) * 2020-05-15 2021-11-19 北京三星通信技术研究有限公司 Image processing method and device, electronic equipment and readable storage medium
CN113781320A (en) * 2021-08-02 2021-12-10 中国科学院深圳先进技术研究院 Image processing method and device, terminal equipment and storage medium
CN113807335A (en) * 2021-01-05 2021-12-17 北京沃东天骏信息技术有限公司 Method, device and storage medium for data enhancement
CN113852759A (en) * 2021-09-24 2021-12-28 豪威科技(武汉)有限公司 Image enhancement method and shooting device

Patent Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110181787A1 (en) * 2010-01-28 2011-07-28 Stmicroelectronics Asia Pacific Pte Ltd Image details preservation and enhancement
CN103065159A (en) * 2013-01-14 2013-04-24 合肥工业大学 Image classification method based on brightness and contrast ratio
US20170193643A1 (en) * 2014-09-30 2017-07-06 Fujifilm Corporation Image processing apparatus, image processing method, program, and recording medium
US20180227509A1 (en) * 2015-08-05 2018-08-09 Wuhan Guide Infrared Co., Ltd. Visible light image and infrared image fusion processing system and fusion method
US20200082367A1 (en) * 2018-09-07 2020-03-12 Bank Of America Corporation Processing System for Providing a Teller Assistant Experience Using Enhanced Reality Interfaces
CN109691980A (en) * 2019-01-07 2019-04-30 哈尔滨理工大学 A kind of diabetic retina image lesion detection method
CN111652806A (en) * 2019-03-15 2020-09-11 上海铼锶信息技术有限公司 Method and system for removing shadow of image
CN109978787A (en) * 2019-03-25 2019-07-05 电子科技大学 A kind of image processing method based on biological vision computation model
WO2021111152A1 (en) * 2019-12-06 2021-06-10 Ibex Innovations Ltd Composition-guided post processing for x-ray images
CN111161181A (en) * 2019-12-26 2020-05-15 深圳市优必选科技股份有限公司 Image data enhancement method, model training method, device and storage medium
CN113674159A (en) * 2020-05-15 2021-11-19 北京三星通信技术研究有限公司 Image processing method and device, electronic equipment and readable storage medium
CN111652817A (en) * 2020-05-28 2020-09-11 大连海事大学 Underwater image sharpening method based on human eye visual perception mechanism
CN112287861A (en) * 2020-11-05 2021-01-29 山东交通学院 Road information enhancement and driving early warning method based on night environment perception
CN112481839A (en) * 2020-11-27 2021-03-12 杭州数迈科技有限公司 Fabric sewing method, system, equipment and storage medium
CN113807335A (en) * 2021-01-05 2021-12-17 北京沃东天骏信息技术有限公司 Method, device and storage medium for data enhancement
CN113139929A (en) * 2021-03-17 2021-07-20 浙江工业大学 Gastrointestinal tract endoscope image preprocessing method comprising information screening and fusion repairing
CN113781320A (en) * 2021-08-02 2021-12-10 中国科学院深圳先进技术研究院 Image processing method and device, terminal equipment and storage medium
CN113852759A (en) * 2021-09-24 2021-12-28 豪威科技(武汉)有限公司 Image enhancement method and shooting device

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
KUANG H等: ""Nighttime Vehicle Detection Based on Bio-inspired Image Enhancement and Weighted Score-Level Feature Fusion"" *
YANG K F等: ""A Biological Vision Inspired Framework for Image Enhancement in Poor VIsibility Conditions"" *
刘富等: ""基于优化MSR的夜间道路图像增强方法"" *
李灿林等: ""一种非均匀照明彩色图像自适应校正方法"" *
王小元等: ""基于视觉感知的低照度图像增强算法"" *

Also Published As

Publication number Publication date
CN114494063B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
US11195258B2 (en) Device and method for automatic image enhancement in vehicles
CN112308799A (en) Offshore road complex environment visibility optimization screen display method based on multiple sensors
CN106683080B (en) A kind of retinal fundus images preprocess method
Ghani et al. Underwater image quality enhancement through Rayleigh-stretching and averaging image planes
CN108257099A (en) A kind of adaptive infrared image enhancing method of view-based access control model contrast resolution
CN101853490A (en) Bionic image restoration method based on human visual characteristics
CN110060221B (en) Bridge vehicle detection method based on unmanned aerial vehicle aerial image
CN109523474A (en) A kind of enhancement method of low-illumination image based on greasy weather degradation model
CN116311254B (en) Image target detection method, system and equipment under severe weather condition
CN116343144B (en) Real-time target detection method integrating visual perception and self-adaptive defogging
CN112287861A (en) Road information enhancement and driving early warning method based on night environment perception
CN112308005A (en) Traffic video significance prediction method based on GAN
Mandal et al. Real-time automotive night-vision system for drivers to inhibit headlight glare of the oncoming vehicles and enhance road visibility
CN112950521B (en) Image defogging method and generator network
CN114494063B (en) Night traffic image enhancement method based on biological vision mechanism
Hmue et al. Image enhancement and quality assessment methods in turbid water: A review article
CN110570469B (en) Intelligent identification method for angle position of automobile picture
CN115439836B (en) Healthy driving assistance method and system based on computer
CN116596792A (en) Inland river foggy scene recovery method, system and equipment for intelligent ship
CN110084118B (en) Pedestrian detection training database establishing method, pedestrian detection method and device
CN106204476A (en) A kind of low-luminance color image enchancing method
CN112926380B (en) Novel underwater laser target intelligent recognition system
CN112836669B (en) Driver distraction driving detection method
CN114359083A (en) High-dynamic thermal infrared image self-adaptive preprocessing method for interference environment
Wang et al. Low-light traffic objects detection for automated vehicles

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
GR01 Patent grant
GR01 Patent grant