Research On Image Magnification Algorithm Based On Sub-pixel Edge Detection

In the study of image magnification, the definition of edges is usually an important basisfor judging quality of image magnification. Image magnification process is divided into twoparts, which are edge of image and interpolation of internal image, the research focuses onthe magnification of edge; in order to improve the accuracy of the edge detection, furtheranalysis on the integer-level edge detection is processed. Then, the result of sub-pixel isachieved. This paper divided the research of image magnification into two sub-part, assub-pixel level edge detection and interpolation algorithm. Focus on these two problems andimprove the algorithm. Finally, the algorithm is realized by software.Firstly, study on the integer-level edge detection operator, and introduce classical edgedetection operators. The simulation of classical edge detection operators are proceed based onthe principle, comparing the simulation results, then choose Canny edge detection algorithm.In allusion to the limitation of Canny algorithm that parameters is fixed, a new method ofscale-adaptive Gaussian filter is proposed as the improved Canny algorithm, and theparameter can be changed with the image. Verify the feasibility of the improved algorithm foreach different type of image information, based on the principle that the filtering scale factoris increasing proportional to the filtering frequency. Verify the superiority of improved Cannyalgorithm, by comparing the detection result of classical and improved Canny.Study on the principle of sub-pixel edge detection based on the completion integer-leveledge detection, divide sub-pixel edge detection into sub-pixel edge location and graycalculation. Work on the basic principles of sub-pixel edge detection and traditional fittingalgorithm. Analyzing the accuracy and complexity of different algorithms, sub-pixel edgelocation method with curve fitting is chose. Aim at the shortage that the edge determinationmethod of second-degree curve fitting does not differentiate different directions, methods ofminimum value and weighting are proposed to improve the original algorithm. Due to theexperiment, the correction of the improved methods are verified. Choose sub-pixelsegmentation algorithm with surface fitting. Combining with the determination of edgedirection in second-degree curve fitting, the sub-pixel gray is decided. The quality evaluationsystem of image segmentation is established, and according to the characteristics of the imagesegmentation, define the Classed Weighted Mean Squared Error as a new parameter toevaluate the quality of image segmentation. Colligate the sub-pixel edge location and graycalculation, a new edge detection method based on second-degree curve fitting and surface segmentation is proposed and used in the image magnification algorithm of this paper.The image magnification is divided into internal image interpolation and edge imageinterpolation. This paper introduces the traditional linear interpolation and non-linearinterpolation algorithm. Analyzing their characters and aim at the smoothness of internalimage, the bilinear interpolation algorithm is selected to magnify the internal image.Traditional sub-pixel edge detection algorithm using integer gray values as calculatedtemplate should not be used because the location and gray of sub-pixel are changed at thesame time, so this paper selects edge interpolation algorithm based on error back-propagationneural network. Low-resolution unit is defined, the image information is classified. Usingsub-pixel edge points as training data to resize the weights detecting by the back-propagationlearning algorithm adaptively, the edge function is achieved. The quality evaluation system ofimage segmentation is established, analyze the quality of the image from subjective visualperception and objective data. Simulate the image magnification algorithm proposed in thispaper and calculate the average gradient value to validate the correction of the new algorithm.This paper uses Visual C++6.0as the development environment. The imagemagnification algorithm based on sub-pixel edge detection proposed is implemented byimage loading, t integer-pixel edge detection, sub-pixel edge detection, image interpolationand image saving and so on, finally displayed through the software interface.