Research On Image Threshold Segmentation In Engineering Draws Based On Artificial Bee Colony Algorithms

Image segmentation is a key problem in the field of computer vision, especially in these fields, such as image processing, analysis and understanding. Image segmentation is a key step from image processing to image analysis. It decomposes an image into regions with different characteristics and extracts the targets from interests; this occupies a very important role in image engineering. The quality of the segmentation will directly affect the performance of the visual system. In the engineering field, engineering drawings are increasing at a speed of tens of thousands every year. These drawings are scanned into the computer and segmented between background and objects by using the image threshold segmentation technology quickly and accurately, then they are stored in the computer with vector form or reuse for drawing images. The way of storage and application has positive significance in improving the business personnel management level and promoting the efficiency of the drawing designers in the engineering application.The key of the problem to be solved is how to obtain reasonable threshold according to the different types of engineering drawing image. Progressive relationship has been proposed for ABC algorithm here, the fitness function is 2-dimensional linear-type cross entropy. Then some research is respectively done on different types of engineering drawing image in threshold segmentation problems.First of all, the fitness function is one-dimensional information entropy; artificial colony algorithm is applied into standard and non-standard engineering drawings with noise for effective separation of target and background. Combined with the segmentation results, the limitation and the direction of the algorithm improvement are summed up by using one-dimensional statistical information as fitness function.Secondly, according to the feature of engineering drawings image, a genetic mechanism binary bee colony algorithm is proposed, which is encoded with binary. And the fitness function is 2-dimensional linear-type cross entropy. Based on the artificial colony algorithm update strategy, the novel one, that is ’Remove the same and Reserve the different’, is put forward, which is similar to the crossover operation of genetic mechanism. Strategy adjustment is similar to mutation of genetic mechanism. The improved algorithm is proved to be convergent by theoretical thought of genetic algorithm, which also has good convergence by different types of function simulation when it is applied to the function library of test function. And further, when it is applied to the images of engineering drawings with noise, from the view of practical application, the algorithm is effective by comparing between algorithms in the image threshold segmentation.Moreover, low SNR engineering drawings, a relative special drawing image, binary code of binary bee colony algorithm based on genetic mechanism involves the code conversion from binary to decimals in the image threshold segmentation, and the characters in update and adjustment strategy of algorithm need to be transferred back to the binary when they are used. At the same time, the random factors exist in the artificial colony algorithm update strategy, which cause the update process of algorithm to move to the direction of the optimal solution or the degradation. Combined with these algorithm characteristics, improved artificial bee colony algorithms based on Tent mapping and quantum are proposed. Improved artificial bee colony algorithm based on Tent mapping is an algorithm that transforms coding to the domain range by Tent mapping of chaos theory. Update strategy of fixed direction is adopted. Local optima is operated by adjust strategy, which has the character that 1 minus the number of 0 and 1 belongs to still the range from 0 to 1, so the diversity of individuals is increasing. This algorithm is proved to be convergent to 1 in probability using the stochastic process and related theory of the Markov chain. And the improved artificial bee colony algorithm, whose encode is based on quantum using the probability amplitude of square in the sinusoidal component mapping to the interval of domain. Update strategy updates quantum bit probability amplitude by adjusting the phase angle in the fixed direction, and which lead the bees of the artificial bee colony algorithm to the current optimal honey eventually; adjust strategy makes use of honey of followed bees complementarily update by referencing quantum non-gate to exchange the chromosomes’ sine and cosine components. According to related property of the Markov chain, the improved algorithm is proved that is still converge to 1.For the above algorithm, firstly, they are examined that they have preferable ability to jump out of local optimum by benchmark functions. When they are applied to the image of standard library and low SNR engineering drawings with noise, by algorithms comparison and data analysis validation, the effect of the improved artificial bee colony algorithm applied to image threshold segmentation is obvious. The quality of segmented image is greatly improved when it is disposed by secondary average smoothing at last.Finally, because of the characteristics of the large scale image in engineering drawings with noise and the limitation of computer ’s software and hardware conditions, the image can’t be read directly. As a result, the large scale image should be divided into several small regions, and local threshold ways on genetic mechanism binary bee colony algorithm, improved artificial bee colony algorithm based on Tent mapping and quantum are adopted in each region respectively, and the fitness function is 2-dimensional linear-type crossing entropy. After the local threshold of each area is obtained, which is obtained the threshold segmentation respectively, the whole image is reassembled by them.The response curve, performance and real data are demonstrated by the above improved artificial bee colony algorithm based on local threshold, the separation between background and objects in large scale image engineering drawings with noise is very effective.