Research On Quaternion-domain Color Image Segmentation Method

With the continuous development of technology and people's growing demand for information,image segmentation as a basis of part of the information processing is very important.However,image segmentation is becoming the image processing bottleneck,although a variety of methods have been proposed,but there is no universal solution that applies to all images.Image segmentation has gained everyone's attention,so it is still one of the hot spot of the study.On the current situation of image segmentation fields,artificial intelligence is very popular,some methods have been found and have got good results.Here we will deeply explore and research the image segmentation.We will thoroughly study of the quaternion exponential moments,quaternion circular harmonic-Fourier Moments,twin support vector machine(TWSVM),PCC,quaternion Wavelet Transform(QWT)and Hidden Markov tree(HMT).Combined with these methods to complete the following works:1.Fist of all,we decomposed the quaternion exponential and extracted the low frequency coefficients as the pixel characteristics.Then,we used two-dimensional Arimoto Entropy and selected the trained samples.Last,we used TWSVM for pixel classification.Using Quaternion exponential moments to extract pixel level feature,without considering the relevance and relation among image components,and raising the image segmentation to the hypercomplex field,the result we obtained is in accordance with human vision.2.Combining the theory of PCC,we extracted the pixel level feature using quaternions on Fourier moment-the circular harmonic and using two-dimensional Tsallis entropy for trained sample selection.The results of all experiments showed that the method has good stability and efficiency especially for the different types of images.3.We proposed the New QWT-HMT image segmentation method.First of all,we decomposed the image by using QWT,then obtained the amplitude and phase of coefficients,and using amplitude and phase to model final HMT.This method considered relations not only among directions,dimensions but also within dimension of inter-QWT coefficients.It sucessfully captured part and texture information of the image by using QWT amplitude and phase achievements and got a pretty good segmentation results.