Study On Interactive Image Segmentation

Image segmentation refers to the low visual description features of the image based on the existence,and the classification of regions with similarity in some features into the same category according to the designed similarity criteria,and the separation of the parts of the image that are of interest from the image,so as to provide possibilities for subsequent image processing.Image segmentation can be divided into interactive image segmentation and unsupervised image segmentation according to whether users participate in the segmentation process.Interactive image segmentation means that in the image segmentation process,the user provides prior information for image segmentation according to the needs of image processing,and then segmentes the desired result according to the prior information provided by the user and the similarity criteria of the design.Although unsupervised image segmentation is relatively simple in operation compared to interactive image segmentation,due to the "morbidity" of unsupervised image segmentation in the application itself.Interactive image segmentation makes up for the shortcomings of unsupervised image segmentation to a certain extent,and has great advantages in many applications.In recent years,with the development of computer vision,image analysis and pattern recognition,interactive image segmentation has become one of the important research fields in image processing.In recent years,the study of interactive image segmentation has become one of the hot research fields explored by many scholars at home and abroad,and many research achievements of great significance to the development of image segmentation have been achieved.In many applied fields,image segmentation has also achieved good achievements.However,the current interactive image segmentation methods still have many problems,and the research on interactive image segmentation needs to be further deepened to overcome the problem that the current image segmentation methods are not ideal for image segmentation due to the diversity and complexity of images.The problems existing in current interactive image segmentation methods mainly include the following:(1)Although existing methods have been used to study the segmentation of segmentation objects with elongated parts,the segmentation effect of segmentation objects with elongated parts is still poor,which needs further study.(2)In image segmentation,the segmentation of images with noise and texture is relatively common.Although there have been more research achievements,especially in the segmentation of natural images,the effect of image segmentation with complex texture and certain noise is not ideal.(3)It can't segment the complex natural image effectively,and it's not suitable for different types of image segmentation.(4)When there are multiple independent same or similar segmentation objects in the image,it is often necessary to provide sufficient seed pixels.When the interaction information is insufficient,it is not possible to effectively segment all the objects in the image.(5)The weak boundary is a common problem in the image.At present,many methods are not very accurate in the segmentation of the weak boundary.In view of the problems existing in the above interactive image segmentation,this paper conducts in-depth research around the random walk image segmentation model,and this paper analyzes the problems existing in the existing interactive image segmentation random walk methods and proposes a multi-layer sub-Markov random walk image segmentation method.In addition,referring to the idea of linear regression in statistical analysis,combined with the related theoretical knowledge of dimensionality reduction and propagation,an interactive image segmentation method for linear regression discriminant is proposed.The main work and innovation achievements of this paper can be summarized as follows:(1)A multi-layer sub-Markov random walk interactive image segmentation algorithm is proposed.By introducing the image patch layer and cooperating with the prior layer to establish the relationship between the image patch layer and the pixel layer,the algorithm achieves the construction of the medium and far distance connection between the pixels,and it enhances the similarity between pairs of pixels,thereby increasing the migration probability of random walker,and helping to achieve the effect of segmenting objects with long and thin sections in the image.In addition,due to the introduction of image patch,it has a certain smoothing effect on the noise and texture in the image,and can have a better segmentation effect on this type of image.Because the transition probability of random walker is increased,it has a good segmentation quality for more complex natural images.The medium and long distance connection between pixels is established by image patch layer,which makes a further development in theory for random walk model.(Corresponds to Chapters 3)(2)An interactive image segmentation algorithm based on dimension reduction and linear regression discriminant is proposed.This algorithm is proposed after the careful analysis of the dimensionality reduction and propagation theory in data classification based on the idea of linear regression in statistical analysis.It uses the dimensionality reduction method to map the high-dimensional feature data describing the image to an effective subspace,removes redundant information,and makes the image segmentation better and more accurate,and by using the propagation theory,the local similarity between the pixels in the image is propagated to the global domain,which makes the similarity between the paired pixels more complete.At the same time,the parameters of the discriminant function effectively extract the essential features of the image,which makes it possible for this method to be applicable to interactive cosegmentation.Because this method uses the propagation theory to propagate the local similarity between pixels to the global domain,this allows better segmentation results for different graph constructions.Since the method uses the discriminant function to segment the image after obtaining the parameters that describe the main features of the image,it can achieve better image segmentation quality for the segmented object with long and thin parts,weak boundary problems,segmentation of multiple independent same or similar segmentation objects.(Corresponds to Chapters 4)(3)For the interactive image segmentation method based on random walk and linear regression discriminant proposed in this paper,in order to verify the effectiveness of the proposed method,a large number of qualitative and quantitative comparison experiments were conducted on natural images,and it is shown that the method proposed in this paper is effective for image segmentation,and it is also shown that the research on interactive image segmentation has certain progressive significance.With regard to the image segmentation method of linear regression discriminant,the experiment not only verifies that the image segmentation effect under different graph construction is effective,but also can effectively realize multiple labels image segmentation.At the same time,a good segmentation effect has been achieved in the interactive co-segmentation experiment,which also provides a new research direction for the research of interactive co-segmentation.(Corresponds to sections 3.4 and 4.4 of the paper)...