Liang Kim Meng

Non-standard Malaysian car license plate recognition

Automated car license plate recognition systems are developed and applied for purpose of facilitating the surveillance, law enforcement, access control and intelligent transportation monitoring with least human intervention. The challenging part of this system lies in accurately recognizing a non-standard plate that consists of various character fonts, character sizes, plate size, orientations, plate and character colors, materials, positions of numeric characters and alphabetic characters. The objective of the paper is to present our developed algorithms that cope with these practical variations in the context of non-standard Malaysian plate from image scene acquisition to optical character recognition. In each standard step of license plate recognition, we consider the processing of the plates under various environmental condition and non-standard formation of characters in real scene. The algorithm is tested with 500 Malaysian car plates in real-time and the result shown that the algorithm remains relatively robust under different conditions and car plate standard.

A New Multi-spectral Fusion Method for Degraded Video Text Frame Enhancement

Text detection and recognition in degraded video is complex and challenging due to lighting effect, sensor and motion blurring. This paper presents a new method that derives multi-spectral images from each input video frame by studying non-linear intensity values in Gray, R, G and B color spaces to increase the contrast of text pixels, which results in four respective multi-spectral images. Then we propose a multiple fusion criteria for the four multi-spectral images to enhance text information in degraded video frames. We propose median operation to obtain a single image from the results of the multiple fusion criteria, which we name fusion-1. We further apply k-means clustering on the fused images obtained by the multiple fusion criteria to classify text clusters, which results in binary images. Then we propose the same median operation to obtain a single image by fusing binary images, which we name fusion-2. We evaluate the enhanced images at fusion-1 and fusion-2 using quality measures, such as Mean Square Error, Peak Signal to Noise Ratio and Structural Symmetry. Furthermore, the enhanced images are validated through text detection and recognition accuracies in video frames to show the effectiveness of enhancement.

A new sharpness based approach for character segmentation in License plate images

Character segmentation from License plate is challenging as it suffers from non-uniform illumination, blur and touching adjacent character due to head light of vehicles effect etc. This paper presents a new approach based on sharpness of character and the space regions for segmentation. We explore Gradient Vector Flow Opposite Direction Pair of pixels for seed points selection. From the seed points, the proposed approach detects probable cuts between character components based on minimum cost path estimation. Then we propose a new sharpness for detecting candidate cuts from the probable cuts based on Laplacian zero crossing points and Sobel-Gradient. The average width of candidate cuts is considered as reference cut to identify the correct cuts between character components. Experimental results on real industry dataset of license plate shows that the proposed approach is robust to touching, blur, poor quality images. Further, comparative study with an existing approach shows that the proposed approach outperforms the existing approach.

Video analytics algorithm for detecting objects crossing lines in specific direction using blob-based analysis

This paper presents a video analytics algorithm for detecting event of objects crossing predetermined line-of-interest in the scene in specific direction. A fast blob-based analysis is formulated to detect the event, combined with the object detection and tracking to detect and tracked the object as motion blobs. Proposed algorithm is tested in real outdoor surveillance environment for 24 hours in 3 days to evaluate the detection accuracies in different scenarios. For comparison, the testing is done against a commercial surveillance system. The results show that the proposed algorithm provides better accuracy in all scenarios, while maintaining real-time processing capacity.