摘 要
粒子群優(yōu)化算法源于鳥群群體運(yùn)動行為的研究，是一種基于種群搜索策略的自適應(yīng)隨機(jī)優(yōu)化算法。作為群智能的典型代表，粒子群優(yōu)化算法已經(jīng)被證明是一種有效的全局優(yōu)化方法，一經(jīng)提出就受到全世界研究者的關(guān)注、重視，目前已經(jīng)被廣泛應(yīng)用于圖像分割、目標(biāo)函數(shù)優(yōu)化、神經(jīng)網(wǎng)絡(luò)訓(xùn)練、模糊控制系統(tǒng)等許多領(lǐng)域，并取得了良好的效果。
圖像分割是目標(biāo)檢測和識別過程中的重要步驟，其目的是將感興趣的區(qū)域從圖像中分割出來，從而為計算機(jī)視覺的后續(xù)處理提供依據(jù)。圖像分割的方法有多種，閾值法因其實現(xiàn)簡單而成為一種有效的圖像分割方法。然而要在直方圖呈多峰分布的復(fù)雜圖像中搜索一個最佳多閾值組合對圖像進(jìn)行分割，它的高耗時性無法滿足實時性的要求，而閾值的準(zhǔn)確確定又是有效分割圖像的關(guān)鍵。因此，快速準(zhǔn)確地搜索到圖像分割的多閾值組合將是問題的難點。要快速和準(zhǔn)確地確定復(fù)雜圖像中的最佳多閾值組合，使分割效果好且滿足實時性的要求，就必須尋求一種高效的算法來解決基于多閾值法的圖像分割問題。
本文在前人工作的基礎(chǔ)上，對粒子群優(yōu)化算法及其在圖像分割中的應(yīng)用進(jìn)行了研究：
（1）為了提高粒子群算法的收斂速度并同時提高算法的全局搜索性能，本文著重研究了兩種新穎的改進(jìn)型粒子群算法。（a）第一種改進(jìn)算法采用相對基初始化粒子種群以獲得更優(yōu)的初始解。該算法為了進(jìn)一步提高收斂速度及精度，當(dāng)群體陷入局部最優(yōu)時，產(chǎn)生相應(yīng)的變異粒子，比較其適應(yīng)度，選取適應(yīng)度高的粒子繼續(xù)優(yōu)化進(jìn)程。通過對不同測試函數(shù)的仿真實驗表明，該算法顯著地提高了粒子群算法的收斂速度和精度。（b）第二種改進(jìn)算法是將粒子群算法與免疫算法相結(jié)合，采用模擬退火機(jī)制對粒子的位置進(jìn)行限制，并用旅行商問題驗證了算法在組合優(yōu)化中的有效性。
（2）將本文改進(jìn)的兩種算法應(yīng)用于基于多閾值法的圖像分割試驗中，實驗表明：該兩種改進(jìn)算法能快速準(zhǔn)確地找到分割閾值的最佳組合，取得好的分割效果且適合多峰直方圖的復(fù)雜圖像。

關(guān)鍵詞 粒子群優(yōu)化算法；圖像分割；變異模型；人工免疫；多閾值









Abstract
Particle swarm optimization algorithm (PSO) is inspired by social behavior of bird flocking or fish schooling.It is a population-based, self-adaptive search optimization technique.As a kind of swarm intelligence, it has been proven to be an effective global optimization method. PSO algorithm has attracted a lot of attention from researchers around the world since it was put forward. It has already been successfully used in many areas, such as image segmentation, function optimization, artificial neural network training, and fuzzy system control.
Image segmentation is regarded as an important step in object examination and recognition.The main goal is to separate objects of interest from an acquired image,so it provides the evidence to the subsequent processing of computer vision.Several methods are proposed from different theoretical point of view for image segmentation. Image threshold segmentation is an effective tool for image segmentation because the simple implemention. However, the problem of time-consuming computation will not meet real-time requirement when we try to search optimum multilevel thresholds on a multimodal histogram of a complex image. But exactly determining those thresholds is the key for effective image segmentation.So it is a difficult problem for us to quickly and exactly search optimum multilevel thresholds for image segmentation.However, to quickly and exactly determine optimum combination of multilevel thresholds, which can segment the image efficiently and meet real-time requirement, we must explore an effective and rapid algorithm to solve the problem of image segmentation based on multilevel thresholds.
Based on the former research,the author studies the improvement of particle swarm algorithm and its application on image segmentation：
Firstly, in order to improve the particle swarm algorithm convergence speed and also improve the global search function algorithm, this paper focuses on rearching two novel improved particle swarm algorithm. (a)The first kind of improved algorithm is adopted Opposition-based Learning initialization particle population, to gain more optimal initial solution. This Algorithm in order to further enhances the convergence speed and precision, when the group into the local optimal, produced the corresponding variation particles, compare their fitness, the selection of the best fitness particle continue to optimize process. According to the different test function of simulation experiment shows that the improved particle swarm algorithm is significantly improved tne algorithm convergence speed and precision. (b)The second kind of improved algorithm combins particle swarm algorithm with immune algorithm and using simulated annealing mechanism of particle position limit, and traveling salesman problem verifies the effectiveness of the combinatorial optimization algorithm.
Secondly, the two improved algorithms are applied to image segmentation experiments, based on multi threshold value. The experiment showed that the two improved algorithms can rapidly and accurately find the best combination of thresholds, obtain good segmentation results and suitable for complex image with multi-modal histogram.

Key words Particle swarm optimization algor..