A Study On The Segmentation For Wetland Bird Image

Wetland is one of the world’s three major ecosystems; it has a very close relation withhuman’s survival, reproduction and development. In recent years, with environmentalawareness gradually rising, people pay more and more attentions to the protection of wetland.Birds are an important indicator of the evaluation of the quality of a wetland; they play animportant role in the wetland protection and monitoring. In recent years, the research on birdimages began to rise.The content which this paper studies is birds image segmentation with the complexwetland environment background. It is the earlier period work of bird recognition, whichprovides the segmentation image that is easier to extract the characteristics for the birdrecognition and automation of wetland monitoring. In view of the characteristics of the imagefor the wetland, this paper uses different methods to pre-segment the image of wetland firstly,then validating the results and forming the final segmentation.The main content and achievement as follow:1Considering the hue information of wetland’s image, to use the method of thresholdsegmentation to pre-segment images in order to attain the zone which may contain birds. Thisis base on experience, which is the rule gained by a large number of statistics on images ofwetland.2Considering the color saturation and brightness of wetland’s image, to reduce image’sinformation dimensions effectively and use FCM algorithm for dealing with the fuzzyinformation. As the wetland image contains sky, river, beach and birds which have the similarcharacteristics, so they are complexity and ambiguity and can not use a single method tosegment. Fuzzy C-means clustering algorithm which has the strict mathematical basis is themost consummate in the number of fuzzy clustering algorithm, and it can handle fuzzyinformation very well.3From the above segmentation results and make the necessary subsequent processing toachieve a better pre-segmentation of the wetland image.4In order to improve effect of bird image segmentation, this paper adds the verificationprocess which verifies the results of the pre-segmentation and removes the incorrectsegmentation. This paper uses the gray level co-occurrence matrix, Gabor features to extractbirds and background texture features, and then use these features to train the BP neuralnetwork. When the neural network training is completed it can be used for bird area detectionand validation, that is, verify the zone of pre-segmentation whether a bird image. Theexperimental results illustrate the segmentation with detection and validation can greatlyimprove the effect of bird image segmentation.