Study On Segmenting And Recognizing Video Object Based On Feature Image Sequences Of K Elements

In all information technology application fields, the feature description, feature holding, searching, recognition and segmentation based on the features of the information data are very important to the design and optimization of the information data collection, transmitting, exchange, receive system. The description of information data based on K feature sequences and The segmentation & recognition of video object(SRVO) based on K feature images sequences of the video object(K-FIS) have not only great theoretical signification, but also great practical signification, which may be applied in many fields such as remote sensing,medicine ,weather, finance ,public security , traffic ,file processing, detection, machine vision , military affairs and so on. The SRVO based on K-FIS can not only segments and recognizes the video object in high efficiency,but also characterizes and compresses the video object in high efficiency, which is very important for multimedia transmitting and search.K feature images means that the video object has K feature images, the status number of each feature images is limited, and the status number of each feature images is probable different from the others. The feature image is correlated on time or space. K feature images may characterize the video object that means if K-FIS is recognized, then the video object is recognized.The SRVO based on K-FIS belongs to the cross-fields of pattern recognition and computer vision, and is extended from the estimation and detection of K parameters to the segmentation and recognition of K feature images.This paper researches and establishes theory model of The SRVO based on K-FIS. This paper also analyzes the characteristics of K-FIS in space relation, information source entropy, status spaces and the relation between the video object and its observer data. This paper also establishes the technical frame of The SRVO based on K-FIS. On the segmentation of feature image sequences, two segmentation methods are proposed in this paper. One is based on the learning of extended characteristic color of the feature images sequences. After calculating the reference clusters of the characteristic color, the reference clusters are fuzzily extended. According to the characteristic color pattern, the object feature images are segmented based on the extended reference clusters. The other method is based on the recognition of the macro feature images according to the transcendent knowledge of the features. Fist, filter the noises based color filter, then locate the feature images cursorily based the characteristic color and the gray texture of the feature images, and then decide if the segmentation is successful according to the transcendent knowledge of the features images. If the segmentation failed, then segment the feature images again based on other parameters or method. On the description and extraction of the visual macro features of the feature image sequences, this paper describes especially the layer object recognition feature of the binary feature images, and proposes a method of extracting the layer object recognition feature. The description and the extraction method of the macro strokes and the macro stroke structures of the character feature images also are given in this paper.On the fuzzy recognition of the video object based on K-FIS, this paper mainly discusses the basic theory of the fuzzy recognition of the video object based on K-FIS include the system, fuzzy set, and fuzzy recognition method. Especially discusses the fuzzy recognition of the video object based on the visual macro strokes and stroke structures feature space of the character images.On the recognition of the feature images based on the BP neuron net group, this paper discusses the recognition feature multilevel model of the K feature images sequences. The feature images can be layered into several layer objects that are described by their style, location and size after the object images are scanned. A new BP-NN group presented in this paper, which consist of a lot...