數(shù)字圖像分割.doc
約64頁DOC格式手機打開展開
數(shù)字圖像分割,摘要圖像分割是一種關(guān)鍵的圖像分析技術(shù)，目的是通過對圖像的分析和研究，將感興趣的區(qū)域或目標(biāo)提取出來。圖像分割是承接圖像處理與圖像分析之間的關(guān)鍵步驟，也是圖像進一步理解的基礎(chǔ)。圖像分割有著篅@さ難芯坷罰恢筆茄芯康娜鵲愫徒溝鬮侍?，几十年来也提畴h聳鄖Ъ頻乃惴?。諒T┓椒ㄋ淙輝諞歡ǔ潭群頭段誚餼雋四承┨囟ǖ奈侍?，但是并?..
內(nèi)容介紹
此文檔由會員 違規(guī)屏蔽12 發(fā)布
摘 要
圖像分割是一種關(guān)鍵的圖像分析技術(shù)，目的是通過對圖像的分析和研究，將感興趣的區(qū)域或目標(biāo)提取出來。圖像分割是承接圖像處理與圖像分析之間的關(guān)鍵步驟，也是圖像進一步理解的基礎(chǔ)。圖像分割有著篅@さ難芯坷罰恢筆茄芯康娜鵲愫徒溝鬮侍?，几十年来也提畴h聳鄖Ъ頻乃惴āＵ廡┓椒ㄋ淙輝諞歡ǔ潭群頭段誚餼雋四承┨囟ǖ奈侍?，但是并不能金q鏊型枷穹指畹奈侍?。并沁b裎掛裁揮幸桓鐾ㄓ玫睦礪劾雌蘭鄯指畹慕峁?，因刺K夥矯嫻難芯棵媼儺磯嗵粽健�
在分割圖像時，由于受到噪聲、光照等污染使得圖像模糊不清，圖像中的細(xì)節(jié)和邊緣等信息無法完全分割出來，本文研究的基于模糊技術(shù)的圖像分割能對模糊降質(zhì)圖像進行有效分割。
本文對數(shù)字圖像分割方法做了系統(tǒng)深入的研究，主要研究工作如下：
1、分析了圖像分割的研究背景，概括了國內(nèi)外的研究現(xiàn)狀和發(fā)展趨勢。
2、深入研究了圖像分割前期的預(yù)處理問題。分析比較了抑制高斯噪聲的均值濾波器和抑制椒鹽噪聲的中值濾波器。在此基礎(chǔ)上針對圖像通常含有的噪聲類型，提出了一種改進的PCNN脈沖耦合神經(jīng)網(wǎng)絡(luò)圖像濾波算法。該方法是通過對每個神經(jīng)元進行受噪類型的判斷，選擇使用不同的去噪濾波器。
3、基于模糊理論的閾值圖像分割方法研究。詳細(xì)分析了圖像分割中的閾值分割方法，針對待分割圖像均或多或少存在模糊不清的情況，結(jié)合模糊理論，將模糊技術(shù)與大津法（Otsu）閾值分割相融合改進了閾值圖像分割方法。實驗表明該方法可以改進因噪聲、光照或其他干擾因素造成的模糊、多目標(biāo)、分割不完全的情況，分割效果明顯改善。
4、改進傳統(tǒng)大津圖像分割方法。為更有效地解決自動選取的閾值偏向于方差較大一類的問題，準(zhǔn)確找到直方圖的谷點位置，更好地分割小對象，改進了傳統(tǒng)大津法。
5、結(jié)合遺傳算法的閾值圖像分割方法研究。為提高分割效率，研究了圖像分割的快速性。利用尋優(yōu)性能較好的智能算法中的遺傳算法來對多閾值尋優(yōu)，實驗結(jié)果表明該方法可以較準(zhǔn)確尋找到一組最優(yōu)解，且耗費時間比模擬退火算法(SA)及窮舉法要少得多。同時與Otsu、最大熵方法比較，結(jié)合遺傳算法的分割方法所耗費的時間要略多一些，但是卻能獲得更高質(zhì)量的圖像分割結(jié)果。折中的結(jié)果是結(jié)合遺傳算法的分割方法能取得更好的效果。
關(guān)鍵詞 圖像分割；濾波去噪；最大類間方差閾值；模糊技術(shù)；遺傳算法
Abstract
Image segmentation is a key image analysis technology, which the purpose is to pick out the regions or objectives of interest through analysis and study of the image. Image segmentation is an important step between image processing and image analysis, it is also the foundation of the further image understanding. Image segmentation has a long research history; it has been a hot research and focus, and thousands of algorithms were proposed. Although these methods have some extent and within a certain range solved some specific problems, cannot solve all the problems of image segmentation. Furthermore, there is not a general theory to eva luate the results of segmentation, so the research in this area faces many challenges.
In the image segmentation, the details, edges and other information in images cannot be completely separated because of image noises, light, etc. So the image segmentation based on fuzzy technology can effectively segment the fuzzy degraded images.
The paper studies the methods of digital image segmentation. The main works include:
1. Analyzed the background of image segmentation, summarized the status of research and the development trend in domestic and foreign.
2. Studied the pretreatment of the pre-segmentation, including compared the mean filter which can suppress Gaussian noise and the median filter which can suppress Salt & Pepper noise. On this basis, considering the types of noise which the images usually contain, proposed an improved PCNN (Pulse Coupled Neural Network) image filter algorithm. The method used different noise filters according to the different types of noise which each neuron was polluted.
3. Research on threshold segmentation based on fuzzy theory. Analyzed the thresholding image segmentation in details, and after studying the fuzzy technology, proposed a new method which combined with fuzzy technology and Otsu thresholding segmentation. It can deal with the images which are fuzzy. The experiment results showed that it can improve the fuzzy, multi-objective, incomplete segmentation situation caused by noises, light and other interference factors.
4. Improved the traditional Otsu, so it can overcome the thresholds trends to larger variance automatically, find the valley of histogram accurately. It can segment the small objective better.
5. Research on threshold segmentation based on genetic algorithm. To improve the segmentation efficiency, studied the image segmentation quickness. It used genetic algorithm which is one of the best optimization algorithms to optimize the multiple thresholds. The experiment results signified the method can find a set of optimal solutions accurately and the time-consuming is much less than the Simulated Annealing (SA) and the exhaustive method. At the same time, this method spent slightly more time than the Otsu and the maximum entropy method, but it can gain better segmentation effect. Compromise is the new segmentation with genetic algorithm can achieve better results.
Keywords image seg..
圖像分割是一種關(guān)鍵的圖像分析技術(shù)，目的是通過對圖像的分析和研究，將感興趣的區(qū)域或目標(biāo)提取出來。圖像分割是承接圖像處理與圖像分析之間的關(guān)鍵步驟，也是圖像進一步理解的基礎(chǔ)。圖像分割有著篅@さ難芯坷罰恢筆茄芯康娜鵲愫徒溝鬮侍?，几十年来也提畴h聳鄖Ъ頻乃惴āＵ廡┓椒ㄋ淙輝諞歡ǔ潭群頭段誚餼雋四承┨囟ǖ奈侍?，但是并不能金q鏊型枷穹指畹奈侍?。并沁b裎掛裁揮幸桓鐾ㄓ玫睦礪劾雌蘭鄯指畹慕峁?，因刺K夥矯嫻難芯棵媼儺磯嗵粽健�
在分割圖像時，由于受到噪聲、光照等污染使得圖像模糊不清，圖像中的細(xì)節(jié)和邊緣等信息無法完全分割出來，本文研究的基于模糊技術(shù)的圖像分割能對模糊降質(zhì)圖像進行有效分割。
本文對數(shù)字圖像分割方法做了系統(tǒng)深入的研究，主要研究工作如下：
1、分析了圖像分割的研究背景，概括了國內(nèi)外的研究現(xiàn)狀和發(fā)展趨勢。
2、深入研究了圖像分割前期的預(yù)處理問題。分析比較了抑制高斯噪聲的均值濾波器和抑制椒鹽噪聲的中值濾波器。在此基礎(chǔ)上針對圖像通常含有的噪聲類型，提出了一種改進的PCNN脈沖耦合神經(jīng)網(wǎng)絡(luò)圖像濾波算法。該方法是通過對每個神經(jīng)元進行受噪類型的判斷，選擇使用不同的去噪濾波器。
3、基于模糊理論的閾值圖像分割方法研究。詳細(xì)分析了圖像分割中的閾值分割方法，針對待分割圖像均或多或少存在模糊不清的情況，結(jié)合模糊理論，將模糊技術(shù)與大津法（Otsu）閾值分割相融合改進了閾值圖像分割方法。實驗表明該方法可以改進因噪聲、光照或其他干擾因素造成的模糊、多目標(biāo)、分割不完全的情況，分割效果明顯改善。
4、改進傳統(tǒng)大津圖像分割方法。為更有效地解決自動選取的閾值偏向于方差較大一類的問題，準(zhǔn)確找到直方圖的谷點位置，更好地分割小對象，改進了傳統(tǒng)大津法。
5、結(jié)合遺傳算法的閾值圖像分割方法研究。為提高分割效率，研究了圖像分割的快速性。利用尋優(yōu)性能較好的智能算法中的遺傳算法來對多閾值尋優(yōu)，實驗結(jié)果表明該方法可以較準(zhǔn)確尋找到一組最優(yōu)解，且耗費時間比模擬退火算法(SA)及窮舉法要少得多。同時與Otsu、最大熵方法比較，結(jié)合遺傳算法的分割方法所耗費的時間要略多一些，但是卻能獲得更高質(zhì)量的圖像分割結(jié)果。折中的結(jié)果是結(jié)合遺傳算法的分割方法能取得更好的效果。
關(guān)鍵詞 圖像分割；濾波去噪；最大類間方差閾值；模糊技術(shù)；遺傳算法
Abstract
Image segmentation is a key image analysis technology, which the purpose is to pick out the regions or objectives of interest through analysis and study of the image. Image segmentation is an important step between image processing and image analysis, it is also the foundation of the further image understanding. Image segmentation has a long research history; it has been a hot research and focus, and thousands of algorithms were proposed. Although these methods have some extent and within a certain range solved some specific problems, cannot solve all the problems of image segmentation. Furthermore, there is not a general theory to eva luate the results of segmentation, so the research in this area faces many challenges.
In the image segmentation, the details, edges and other information in images cannot be completely separated because of image noises, light, etc. So the image segmentation based on fuzzy technology can effectively segment the fuzzy degraded images.
The paper studies the methods of digital image segmentation. The main works include:
1. Analyzed the background of image segmentation, summarized the status of research and the development trend in domestic and foreign.
2. Studied the pretreatment of the pre-segmentation, including compared the mean filter which can suppress Gaussian noise and the median filter which can suppress Salt & Pepper noise. On this basis, considering the types of noise which the images usually contain, proposed an improved PCNN (Pulse Coupled Neural Network) image filter algorithm. The method used different noise filters according to the different types of noise which each neuron was polluted.
3. Research on threshold segmentation based on fuzzy theory. Analyzed the thresholding image segmentation in details, and after studying the fuzzy technology, proposed a new method which combined with fuzzy technology and Otsu thresholding segmentation. It can deal with the images which are fuzzy. The experiment results showed that it can improve the fuzzy, multi-objective, incomplete segmentation situation caused by noises, light and other interference factors.
4. Improved the traditional Otsu, so it can overcome the thresholds trends to larger variance automatically, find the valley of histogram accurately. It can segment the small objective better.
5. Research on threshold segmentation based on genetic algorithm. To improve the segmentation efficiency, studied the image segmentation quickness. It used genetic algorithm which is one of the best optimization algorithms to optimize the multiple thresholds. The experiment results signified the method can find a set of optimal solutions accurately and the time-consuming is much less than the Simulated Annealing (SA) and the exhaustive method. At the same time, this method spent slightly more time than the Otsu and the maximum entropy method, but it can gain better segmentation effect. Compromise is the new segmentation with genetic algorithm can achieve better results.
Keywords image seg..
