Research On Engineering Drawings Denoising Based On Wavelet Analysis

Engineering drawings are precise auxiliary tool in project. With the development ofmodern technology, more and more paper drawings are scanned into digital drawings. Andthe information can be made, modified and added and other operations via computer, whichthe efficiency and quality of preservation drawings are greatly improved. However, thedigitized drawings often are mixed with some noise for the paper drawings are beingcontaminated easily by themself, or there may exsit some thermal noise generated by themachine running in the scanning process. As the engineering drawings are mainly composedof points, lines and other small symbols, the subsequent processing of engineering drawingsand conservation value are inevitably affected by the presence of noise.Like the methods of ordinary image denoising, the methods of engineering drawingsincluding traditional methods which are Airspace domain method, Frequency domain methodand modern wavelet analysis method. The noise is generally removed by the cost ofsmoothing the image of traditional method, and thus there is a shortcoming of missing imagedetail. The wavelet analysis method as the emerging modern mathematical analysis methods,its good performance of multi-resolution analysis, time-frequency localization properties andMallat algorithm, making it be one of the most widely used methods in the field of imageprocessing, and its image denoising process is no exception, the wavelet denoising methodhas been widely studied and applied for its good denoising effect and easy to beimplemented,The inadequate of the wavelet thresholding method which is used for engineeringdrawings denoising was mainly researched. The main points as follows:Firstly, the advantages and disadvantages among the wavelet modulus maxima method,wavelet correlation coefficient method and wavelet thresholding method were researched, forthe method of wavelet modulus maxima and wavelet correlation coefficient has theshortcomings of difficult and complicated. And the wavelet threshold denoising algorithmwas chosen as the actual engineering drawings for its simple and easy to be implemented.Secondly, for the defects that the image can only be decomposed into a finite orientationby the wavelet transformation, and multi-directional of image can not be better characterized,the directional filter will be applied to the method of wavelet threshold in this paper, namely,the hybrid wavelet-directional filter bank (HWD) instead of the mere wavelet transformwasused in paper to retain more image information; For the phenomenon of amplitude crossexists between image wavelet coefficients and noise wavelet coefficients, and the method of wavelet thresholding was restricted by this shortcoming. Based on the principle of thewavelet thresholding method was analyzed, the fuzzy sets formed by pixel was converted tothe pixel membership function that whether the pixel belong to the image was characterized,and the membership weights of HWD was built by the membership function. Not only thenumber of image direction was increased but also the gap of coefficients amplitude betweenthe image and the noise was increased by the improved HWD, and the wavelet coefficients ofnoise and image was distributed on both sides of the threshold as much as possible. Thewavelet coefficients of noise were set to zero maximum with losing the image coefficients atleast by this improved HWD.Thirdly, the shortcoming that not considereding the wavelet coefficients below thethreshold which may contain image detail and was set to zero blindly of raditional thresholdfunction. The Garrote threshold function was proposed in this paper based on an attenuationmethod with three adjustment factors, which was used to increase the performance andflexibility of the threshold function. The way that the wavelet coefficients below the thresholdwas set to zero gradually and quickly from the threshold point to the zero position was takenby the threshold function. And the wavelet coefficients near the lower threshold which maycontain the information of image were retained, so as to more image details werepreserved.The convergence and continuity of this improved threshold function was proved.Fourth, for the difficult that the threshold and three adjustment factor parameter wereharder selected in improved Garrote threshold function, and the optimal value which madethe best denoising effect often could not be obtained easily, so the threshold and theadjustment factor were selected uniformly and optimally by simulated annealing algorithmfor its effectively escaping from the local optimal solution in this paper. The purpose is thatthe complexity of parameter selection in the improved Garrote threshold function wasreduced, and the denoising speed and performance of wavelet thresholding were improved;And K-Means algorithm was used to help simulated annealing algorithm to jump out thelocal optimal solution as soon as possible, so the solving speed of the simulated annealingalgorithm could be improved, thereby the image processing speed could be increased alsowhile using the wavelet thresholding method.Fifth, the noisy type and quantity of drawing which impact on the denoising effect byusing the wavelet threshold were analyzed. The appropriate type and intensity of noise for thewavelet thresholding was selected in MATLAB, and performance of wavelet thresholdingalgorithm improved by each point were analyzed and researched too. The waveletthresholding method combined with the various improved points was used to denoise the actual drawings of difference, and the result was compared and analyzed with otherstraditional algorithm.