Aerial Image Enhancement And Road Extraction And Analysis

With the rapid development of information science and technology,the information of aerial images has become one of the most important sources to acquire large the amount of ground observation data.This dissertation is for the image enhancement and for the road detection and analysis in the aerial images of low quality.The main layout of the dissertation is as follows:1.The power transformation and scale changed Retinex algorithm.Through analyzing the development of the Retinex theory in the de-hazing images,the new algorithm based on improved Retinex is proposed for the images with the large difference of depth information.First,the power transformation is used to enhance dark areas and compress the dynamic range of an image,meanwhile,the non-linear transformation is applied to suppress the high light areas;second,the Gauss filtering scale of Retinex is estimated according to the calculated transmission map;at last,the illumination component is obtained by the convolution of Gauss function with the original image,and the enhancement image is got by exponential transform after subtracting the illumination components from the original image.2.The improved fractional differential enhancement algorithm.On the basis of elaborating the definition and characteristics of fractional differential,and on the study of construction and selection of the G-L fractional differential operator,the improved fractional differential image enhancement algorithm is studied according to the low difference of aerial image depth information.Through the comparison with other algorithms,the sudied fractional differential operator can make up for the deficiencies of the traditional image enhancement operator.3.Improved Canny edge detection algorithm.The enhanced image is segmented with the improved Canny edge detection operator,in which,the high and low thresholds are determined by two different math methods for different images,and it can automatically threshold the image into a binary edge image.Subsequently,the linear and curved road segments are regulated by the Hough line transform and extracted based on several thresholds of road size and shapes,in which a number of morphological operators are used,such as thinning(skeleton),junction detection,and endpoint detection etc.In experiments,a number of vague aerial images with bad uniformity are selected for testing.Similarity and discontinuation-based algorithms,such as Otsu thresholding,merge and split,edge detection-based algorithms,and the graph-based algorithm are compared with the new algorithm.The experiment and comparison results show that the studied algorithm can detect most road edges with fewer disturb elements and trace roads with good quality.4.Valley edge detection algorithm.The algorithm firstly transforms the color image into a grey level image without lost road information,then the image is smoothed by using a Guacian filter,and enhanced by using a fractional differential operator for sharpening roads,and finally a proposed valley edge detection algorithm is applied to detect the roads(lines and curves).For each detecting point in the grey level image,the algorithm detects lines(3-4 pixels)in four different directions,to determine if the detecting point is the candidate of the valley edge point.After that,the extracted lines are smoothed by using the thresholds of the line lengths and orientations,the gaps in curves or lines are connected based on artificial intelligent functions,and the noised lines or pixels are removed.If roads are thick,the image can be shrank to the road width of 4~8 pixels,then,the detecting result is returned to the original image.Experiments showed that this algorithm has the better road detection results,especially for the thin and fuzzy roads.5.Minimum circumscribed rectangle.The shape parameter of the linear target is determined and analyzed through a new algorithm of the minimum circumscribed rectangle based on image 1st,and 2nd moments.The center of an object in the binary image can be obtained by using 1st moments,then the major direction through the center can be got by applying 2nd moments,in this case,we got both maximum axis and minimum axis through the center,and algorithm moves the axes parallel away the center until cross object boundary in four directions respectively,the four lines construct a rectangle which is the minimum circumscribed rectangle.Due to the linear target shape is irregular,the linear target length of the calculation precision is affected.In order to improve the accuracy of measurement and analysis,a road is divided into different segments according to curvature differences,and then the segments are measured with the minimum circumscribed rectangles.The shape parameter of the linear target is determined and analyzed.The experimental results show that the algorithm has the good performance.