Research Of Image Semantics Underlying Feature Extraction

It is easy to produce the "semantic gap" between low-level visual features and high-level semantics of the image for the image is to use their prior knowledge to the process of reasoning image semantics. In order to reduce the "semantic gap" from the perspective of the underlying features rough set theory is introduced into the extraction of the semantics of the underlying characteristics of the image, and ultimately the underlying feature set was validated by support vector machine (SVM).We can use such as color, texture, shape, or a combination of features to represent the semantic content of images in the underlying characteristics of the image semantics extraction, these features can be independent, objective manner to obtain information directly from the image. Typical feature extraction method include the great-very small and the forward sequential, the sequential branch and bound search algorithm, but these methods do not have the knowledge systems and classification of closely linked, they can not be found and reasoned the relationship between various data characteristics, but also can not effectively deal with imprecise, inconsistent, incomplete information, and discover hidden knowledge, revealing the potential the law. Rough set theory does not require pre-given the number of description of certain characteristics or attributes, such as statistical probability distribution, or membership or probability values in the fuzzy set theory, but start directly from the description of the given problem setting to determine the approximate of the domain of the given problem, and to find the problem in the inner law.At first, This article analyzes the semantic level model, and introduce some commonly used methods for extracting image semantics. On the basis of a separate analysis of the color, texture, shape and other characteristics of the image, we introduce the rough set into the selection of the effective low-level feature set. This paper focuses on the application of the reduction algorithm of rough set theory in image semantic retrieval. We applied knowledge reduction to the feature extraction of image semantics by the reduction algorithm of rough set theory. The decision table shows a special and important knowledge representation system, under the premise of the classification is not affected by building decision-making table, reduction can effectively reduce the dimension of the SVM training samples. Finally, on the basis of the reduction, the using of optimization algorithms for SVM kernel parameter optimization, makes the semantic recognition rate reached a high level. The paper first using K-means clustering algorithm to discrete for the rough set can only deal with the concept of a class or qualitative data object in the knowledge reduction. The article use the most favorable image semantic identification characteristics by the rough set which reduced the sample characteristics. Finally, we verify the validity of this method to extract the semantics of the underlying features by SVM.