Due to increase in the human activity and industrial production the amount of atmospheric particulate matter has got increased which leads to the increase in the fog or haze. The particulate matter in the air in the form of fog or haze leads to limited visibility. For many computer vision applications this poor visibility has become very difficult problem. The visibility of scene degrades because of the light reflects from the atmosphere and objects are absorbed and scattered by the aerosols from the atmosphere Due to haze the scene captured by the system suffers from low contrast, poor visibility low luminance, dimmed brightness and distorted color.
Haze degrades the image quality so it is annoying problem for the photographers. Removing haze from the captured images is important in outdoor computer system application as haze affects the reliability of many applications like tracking, object detection, outdoor surveillances and aerial images.
Also while driving haze results in the limited visibility and it becomes difficult to detect to detect the edge of road which may leads to the accidents. Therefore improvement in the visibility also caused as dehazing and road edge detection is an inevitable task.
Here for dehazing image RGB separation, histographic equalization used and for the edge detection canny and prewitt algorithm are used.
“Dehazing Technique for Natural Scene Image Based on Color Analysis and Restoration with Road Edge Detection.”
-By Kyamelia Roy, Saurav Kumar, 2017 IEEE Conference
It is about the dehazing techanique for foggy images based on fast restoration of color by color analysis .Due to fog the captured image of outdoor scene degrades and results in poor visibility and as the image details suffers the road egde of the image also degrades so it is extended to road edge detection. Here canny technique is used for edge detection. And for boundary tracing Hough transform technique is used.
“Development of image dehazing system.”
-By Amruta Deshmukh ,Satbir Singh,2016 IEEE Conference
It explains the method of restoring the foggy image and the hardware implementation is explained. Accidents of the vehicles occur frequently in the foggy condition. So to avoid that, the developed dehazing technique mean channel guided algorithm is used. Also complete hardware is presented which is low at cost and uses less power. It uses portable processing core raspberry pi and display screen.
“Comparision of Various Edge Detection Techniques”
-By Simranjit Singh, Rakesh Singh,2015 IEEE Conference
This paper compares various edge detection techniques. The important feature of image like lines, curves and corners can be extracted using edge detection. The abrupt change in the intensity values is signified by the edge in the image. In this paper all the edge detection techniques are compared on the basis of MSE and RMSE also on the bases of PSNR.
This paper gives information about different types of edges and gives the comparisons between canny, prewitt and sobel edge detection techniques.
“Visibility Enhancement in the Foggy Environment Based on Color Analysis”
-By Da-Jinn Wang, Chao-Ho Chen, Tsong-Yi Chen, Wen-Wei Tsai, 2009 IEEE Conference.
Due to fog the captured image visibility degrades. In this paper the technique is based upon color analysis and the atmospheric scattering exponential contrast enhancement technique for enhancing the visibility of the foggy image by using RGB color and tri-stimulus values. Then to increase the brightness of the dehazed image, weighted redistribution of probability density is used. And finally to improve the contrast histograame equalization technique is used.
3.1 PROBLEM STATEMENT
Air pollution is one of the most serious problems. Introduction of chemical particulate matter or biological material in the atmosphere causes harm or discomfort to human or other living things. Haze is one of the basic forms of air pollution which is the major cause of reduced visibility. When sunlight encounters with the tiny particulate matter in the air, fog or haze occurs. Haze reduces the color and clarity of what we see.
Many times clear visibility is required in the input image of the outdoor computer vision application like object recognition, remote sensing image, object detection, tracking etc. but unfortunately this is not always true in many situation particularly when haze occurs frequently.
Due to haze the captured image suffers from the low contrast, poor visibility and distorted color. Also as the image details suffer due to fog, the road edges of the outdoor scene captured degrades significantly.
Thus to solve these problems dehazing of image captured is required along with the image details suffer due to haze, the road edge of captured outdoor scene degrades road edge detection.
Read the input image I.
STEP 2 :
Perform RGB separation
Enhance the image and compliment it.
Add and subtract some constant to adjust darker and lighter region contrast respectively.
Get complemented image.
Subtract resultant image from original enhanced image.
Perform edge detection.
4.2 BLOCK DIAGRAME
Fig.5.1: Block Diagram
4.2.1 RGB Separations:
The RGB image is stored in form of m by n by 3 data array. The RGB image is also called as true color image. Each image consists of pixels and color of a pixel is a combination of three primary colors red, green and blue. The color of pixel is determined by the intensities of color at the pixel location.
The RGB separation will create the three separate color planes containing uninterrupted areas of red, green and blue color. After RGB separation the white part in the planes indicates highest concentration of pure color and the gray indicates the mix of two remaining color values and black indicates no values of true color present.
4.2.2 HISTOGRAM EQAULIZATON
Using the Histographic equalization the dynamic range and contrast of the image is modified by altering the intensity of the image. By employing monotonic, non linear mapping histogram equalization reassigns the intensity values in the image so that the output image contains uniform distribution of the intensity.
Using this technique the lower contrast areas are converted into higher contrast areas. To accomplish this histogram equalization spreads out the most frequent intensity values. The images with much higher color depth than the palette size like continuous data or 16 bit gray scale image will give the best results.
A binary image can be created by using thresholding and it is the simplest method of image segmentation. By turning all the pixels above some thresholding value to one and all the pixels below some threshold value to zero, thresholding creates the binary image from the gray scale.
4. EDGE DETECTION
Edge detection is the technique which finds the points in the digital image where the brightness changes sharply or the image has discontinuities. It is organized into set of curved line segments termed edges where the image brightness changes sharply.
The two types of edge detection techniques are canny edge detection and prewitt edge detection. Canny edge detector algorithm is also called as multi stage algorithm as it detects a wide range of edges in the images.
It reduces amount of the data to be processed for edge detection. And prewitt edge detection is a discrete differentiation operator. Usually prewitt operator gives the direction of largest possible increase from light to dark and rate of change by calculating the gradient of the image intensity at each point.
The first step in the dehazing an image is image restore the lost contrast of the dehaze image. The RGB image consist of pixels in array from each pixel defines red, green and blue components. Each pixel color of an image is formed by the combination of three primary colors red, green, and blue. The decimal code of all colors range between (0,0,0)black and white(255,255,255).the true decimal code for true red is(255,0,0) for blue is (0,0,255) and green is (0,255,-0).for dehazing image first the red green and blue color are separated from image by RGB separated
The contrast of degraded image is enhanced by histrogram equalization. Usually the contrast of the image is increased by this method and intensities can be distributed in a better way on the histogram. In this method the most frequent intensity values are used.
Then to get the complementary RGB image each pixel of red green and blue plane is subtracted by the maximum pixel value which is 255. The complimentary RGB image appears to be darker at low contrast region and light at the high contrast region. Then the contrast of the image is adjusted by adding some contrast value at adding some constant value at the darker region where as subtracting constant value from the lighter region. Both the constant value for the red plane and green plane is 90 and for blue plane is 120.
After this step all three RGB complemented images are concatenated to obtain complemented dehazed image. To obtain final dehazed output image, the complemented image is subtracted from the original enhanced image.
After getting dehazed output, road edge detection is performed by using two algorithms in MATLAB.
I. Canny edge detection
II. Prewitt Edge detection
Canny edge detection uses two thresholds to find strong and weak edges. Derivative of Gaussian filter is used to calculate gradient. Basically canny edge detection is divided into five parts:
First in order to remove the noise Gaussian filter is applied.
Second step is to find the intensity gradient of the image.
To get rid of spurious response to edge detection, apply non maximum suppression.
To determine potential edged apply double threshold.
By using hysteresis track the edges.
In prewitt edge detection difference between corresponding pixel intensities of image is used to calculate edges. Technically prewitt edge detection is nothing but discrete differentiation operator which computes approximation of gradient of intensity functions of an image.
Fig.7.1: Color analysis
The fig.7.1 show the image color analysis. Here images in the first column are the RGB separate planes . Images in the second column are the enhanced red ,green and blue plane. And images in the third column are the complemented images of red , green and blur plane.
Fig.7.2: Red plane
The Fig.7.2 shows the red plane image after RGB seperation.Here the true red color pixels are seperated from the RGB color image.
Fig.7.3: Green plane
The Fig.7.3 shows the green plane image after RGB seperation.Here the true green color pixels are seperated from the RGB color image.
Fig.7.4: Blue plane
The Fig.7.4 shows the blue plane image after RGB seperation.Here the true blue color pixels are seperated from the RGB color image.
Fig.7.5: Complemented image
The Fig.7.5 shows the RGB image after dehazing and complementing. To get this output image the complemented red, green and blue plane images are comcatinated.
Fig.7.6: Dehazing output
The Fig. 7.6 shows the dehazing output image.
Fig.7.7: red plane histogram
The Fig. 7.7 shows the histogram of the red plane. It gives the intensity distribution of red plane pixels.
Fig.7.8: Green plane histogram
The Fig. 7.8 shows the histogram of the green plane. It gives the intensity distribution of green plane pixels.
Fig.7.9: Blue plane histogram
The Fig. 7.9 shows the histogram of the blue plane. It gives the intensity distribution of blue plane pixels.
Fig.7.10: Boundary detection
The Fig.7.10 shows the boundary before edge detection.
Fig.7.11: Road edge detection
The Fig.7.11 shows road edge detection. First image is the original image. Second image is the canny edge detection output. And third image is the prewitt based road edge detected output image.
Due to haze the captured image of the outdoor scene suffers from low contrast, color distortion and poor visibility. Also as the image details suffer due to the fog the road edges of captured image degrades. Hence the proposed system concentrates on the dehazing the captured image and road edge detection,
The proposed system is implemented on personal computer and MATLAB 2015Ra software for execution. By using image color analysis and histogram equalization the dehazed output is obtained. Using canny edge detection and prewitt edge detection road edge detection is performed.