Liu Jiafeng

The design of HMM-based banknote recognition system

The banknote recognition system based on hidden Markov models (HMM) is proposed. It is based on the empirical risk minimization (ERM) principle. Image preprocessing includes brightness equalization and tilt correction. In order to satisfy the high speed and reliability of the banknote processing system, the grid segmentation is used for features extraction. Analyze the experimental data and determine the number of states, iterations, and Gaussian components. The proposed banknote recognition system can be applied to classify any kinds of banknotes. More than 16,000 RMB samples are sampled by CIS (Contact Image Sensor) with 25dpi. Experimental results show that the proposed method obtained higher recognition rate than ANN and SVM.

Multi-object segmentation based on pulse coupled neural network

This paper introduces an approach for image segmentation by using pulse coupled neural network (PCNN), based on the phenomena of synchronous pulse bursts in the animal visual cortexes. The synchronous bursts of neurons with different input were generated in the proposed PCNN model to realize the multi-object segmentation. The criterion to automatically choose the dominant parameter (the linking strength beta), which determines the synchronous-burst stimulus range, was described in order to stimulate its application in automatic image segmentation. Segmentations on several types of image are implemented with the proposed method and the experimental results demonstrate its validity.

Visual tracking via weighted sparse representation

Recently, sparse representation has been used in visual tracking, and related trackers have emerged. However, such sparse representation is not stable and has the potential to represent a candidate with dissimilar target templates. Therefore, a new tracker based weighted sparse representation (WSRT) is proposed. Specifically, to represent a candidate, each target template is weighted according to its similarity to the candidate. The bigger the similarity is, the bigger the probability of the target template to be chosen will be. The proposed tracker chooses the similar target templates to represent each candidate and reflects the locality structure between the candidate and target templates. Experimental results show that the proposed tracker has excellent performance.