Research On The Context Information Based Object Recognition

With the rapid development of multimedia technology in the research field of computer science and its widespread applications in aspects such as military, medical science, transportation, computer vision is gradually becoming a research field with great concern, which also facilitated related subjects such as machine learning, artificial intelligence, robotics to do further researching. Around the research field of computer vision, object recognition is admitted to be one of the most challenging research areas. Generally, tasks in object recognition can be unfolded into two aspects, object representation and recognition. It involves in many comprehensive subjects including image processing, machine learning and pattern recognition etc. At the same time, it is also one of the most significant research areas since it plays an important role in industrial application. All of these make object recognition a research focus in recent years.The main research task in this dissertation is to recognize object in digital image with analysis on the use of prior knowledge during this procedure. Main content of this paper can be summarized as follow:We firstly introduce the basic theorems and approaches in this research area while analyzing the research status and challenges. Due to the problem that the visual features based traditional object recognition methods cannot perform well in a more complex circumstance, we then try to propose a novel approach which combines object recognition with the prior relations. Finally, we compare the proposed object recognition method to traditional method in order to verify the effectiveness of the proposed method.The main stages of the proposed algorithm include two aspects which are called training phase and testing phase. During the training stage, structured presentation of the prior relations is applied through a hybrid graph which contains image similar sub-graph, semantic similar sub-graph and the relations between the two sub-graphs. A random walk model is then constructed according to this hybrid graph. During the recognition stage, a new testing image node is added to the random walk model. Then the relations between this node and the nodes in the random walk model are calculated. At last, random walks which start from the testing image node are performed at the random walk model. The probability rank provided by the result of random walks will serve as the recognition result of the testing image. Experimental results illustrate the validity and higher recognition performance of the proposed method.