Research On The Microscopic Image Segmentation Of No-setae Algae Species Based On Salient Region Detection And Watershed

Phytoplankton is an important part of aquatic organism. Water quality assessment,monitoring harmful algal blooms (HABs) and aquatic ecosystem research are allbased on the research of the qualitative and quantitative of the phytoplankton. Withthe progress of the study, it has been emerged in large numbers of methods, such asflow cytometry, molecular techniques and microscopic image analysis and so on.Among these methods, microscopic image analysis has been widely used as itslabor-saving, fast to implement and low cost. In this paper, a method combined salientregion detection and watershed algorithm is proposed to segment cells in no-setaealgae species microscopic image, and then applied to algae identification and counting.The main work of the thesis:1. Introduction of visual attention and its applications in microscopic images. Firstintroduce neurobiology of visual attention and then analyse three classicalmethods of visual attention. Based on the characteristics of phytoplanktonmicroscopic image, this paper uses an improved IG method to detect salientregion of no-setae algae species microscopic image.2. Microscopic image segmentation of no-setae algae species. This thesis researcheson an improved watershed algorithm based on salient region detection andcharacteristics of microscopic image. For single cell image, maximum areaextraction is used to select markers, the results are compared with other5imagesegmentation methods and made a quantitative analysis; as for multiple cellsimage, Otsu is utilized to select markers, visual effect and quantitative analysis areused to illustrate the result.3. The applications of image segmentation. Algae species identification is aimed atsingle-cell image: shape factors and moment invariant features are extracted as a 19-dimension feature vector, SVM is used as a classifier. The experiment hastrained with1800images covered18spieces, and600images are used to verifythe effect, the recognition rate is82.31%. Connected components are used tocount cells in multiple cells image.93images including703cells are applied inthe experiment and the experiment result is763,79.3%are the real cells whichtake up86.63%of the703cells.The research in this thesis can not only gets a good segment result, but alsospreads to the algae identification system and counting system, therefore hasimportant significances in applications.