Image spam has become a real threat to email communication these days, since most prevalent content based spam filters can not efficiently detect them out, even when the latest OCR techniques are employed, spammers could compromise the system easily through text distortion and other obscuring skills. In this paper, we propose a novel and efficient image modeling approach for spam image classification, this content based statistical model does not rely on the availability of text information embedded in the image files, so that it is robust to obfuscations. Experimental results show that the proposed method can perform with good accuracy in practice.
The diversity of an ensemble is deemed to be a key factor which determines performance in ensemble learning. A variety of approaches have been advanced to quantify diversity by analyzing the prediction of classification which relies on the validation set. This paper proposes a new method how to measure diversity and ensemble for linear kernel Support Vector Machine, which is based on the characteristic parameters of Support Vector Machine. The new method is proved to achieve better performance than the traditional measures of diversity such as Discrepancy method. Further research on relationship between diversity and accuracy is conducted by the method.
Formal semantics of components are foundations for rigorous analyzing and reasoning about the composition process and its correctness. According to the notion of software contract, components interaction patterns and composition process patterns, formal semantics of components are proposed. With this basis and inspired by typing system and process construction methods of -calculus, a typing framework for the composition are proposed. Additionally, based on the operational semantics, a formal model of component is suggested. Then, transition rules about component composition are introduced based on the -calculus typing rules. At last, the feasibility and validity of the proposed composition method are confirmed by the results of composition experiments on a composition platform FSCC.
In order to reduce encoding complexity of H.264/AVC, and improve the coding speed, an adaptive fast coding block mode selection algorithm is proposed. The threshold information is used to select the skip mode effectively and skip unnecessary block mode selection according to the information of the coded frames and the coded blocks. The optimal coding block mode for current block is adaptively selected. The experimental results show that, under the cost that bit rate increased 0.41 percent and the image peak signal to noise ratio reduced 0.034 dB in average, the encoding speed increases 50.40 percent in average.
Analyzing the task management and scheduling algorithm of embedded systems, we present an improved embedded system scheduling algorithm and increase of time slice cycle algorithm. The embedded system is improved upon a real-time system, in which assignment priority scheduling is primary and time slice cycle scheduling is secondary. Through the application of collision detection algorithm and path planning algorithm in the system, it shows that the improved embedded system can well meet the requirement, with good usability.
