| Neural stem cells are kind of cells, derived from neural system. These cells have so good plasticity that can be artificially built by means of gene engineering or cell engineering and some other modern techniques in order to make them directional differentiation and to build up a kind of replacement of impairment in nerve center. Aided by the imaging analysis technique of digital image in computing science and the technique of high speed processing in signal and information, the segmentation of neuron stem cells becomes more popular in the world.The original images analyzed in this dissertation are based on the time lapse images sequence which are acquired from cultured neuron stem cells in vitro captured by con-focal microscopy. Based on mathematical morphology watershed segmentation, improved watershed algorithm is proposed here. Nearest-neighbors Interpolation,Otsu's threshold and region merging are first applied to preprocess the original images. Then the binary images are acquired. An iterative morphology operation is then applied to segment all the clustered cells. Each component in segmented image is analyzed based on their feature characteristics. Noises, under and over segmentation, including splitting cells are extracted in order to be processed respectively. Chamfer 3-3-3 transform algorithm is applied to deal with under-segmentation cells. After distance transform, coarse erosion structures are applied firstly and fine erosion structures are applied secondly. Then we use the marker-controlled watershed algorithm for the cells under segmented. In order to improve the rate of segmentation of the whole images sequence after an interactive correction to the starting frame in sequence, inter-frame information is used.The above algorithms are applied to the separate the clustered objects of the binary image segmentation. The experimental results show that better and more robust performance can be obtained by our proposed method. |
PDF Full Text Request