Ukrit Watchareeruetai

A computer vision based vehicle detection and counting system

A vehicle detection and counting system plays an important role in an intelligent transportation system, especially for traffic management. This paper proposes a video-based method for vehicle detection and counting system based on computer vision technology. The proposed method uses background subtraction technique to find foreground objects in a video sequence. In order to detect moving vehicles more accurately, several computer vision techniques, including thresholding, hole filling and adaptive morphology operations, are then applied. Finally, vehicle counting is done by using a virtual detection zone. Experimental results show that the accuracy of the proposed vehicle counting system is around 96%.

Leaf identification using apical and basal features

This paper proposes a method that extracts local features, i.e., angle patterns, around the apex and base of a leaf. The proposed method only requires leaf contour and the location of apex and base as inputs. Starting from an origin point, which can be either the apex or base, the contour is tracked in two directions, i.e., leftward and rightward, and then sampled at five different distances from the origin point. The angle formed by the origin and two sampled points, at each distance, is then calculated. Altogether, 10 angle features, five from the apex and five from the base, are obtained. These features are invariant to translation, rotation, and scaling. In addition, this paper also aims to measure the effectiveness of the proposed apical and basal features. In the experiment, two sets of features are compared. The first set includes 12 shape descriptors while the second set includes not only the 12 shape descriptors but also the proposed features. By using support vector machine as a classifier, leaf identification has been done by using the two sets of features. Experimental results indicate that the use of apical and basal features can significantly improve the accuracy of leaf identification.

Detection of fibrosis in liver biopsy images by using Bayesian classifier

In this paper, an image-processing-based method designed to detect fibrosis in liver biopsy images is proposed. The proposed method first enhances the color difference between liver tissue and fibrosis areas. Then, a low-pass filtering is applied to each color band to reduce noise. In order to calculate the percentage of fibrosis against total liver tissue, the background area, i.e. empty slide area, is detected. Next, Bayesian classifier is used to separate fibrosis from liver tissue based on the color information. Finally, the proportion of the fibrosis area to the tissue area is computed. Experimental results show that the proposed method can estimate and detect fibrosis in the liver biopsy images with the classification accuracy of 91.42%. In addition, the average difference between the percentage of fibrosis obtained from the proposed method and that in ground truth images is 2.29 points.

Redundancies in linear GP, canonical transformation, and its exploitation: a demonstration on image feature synthesis

This paper concerns redundancies in representation of linear genetic programming (GP). We identify the causes of redundancies in linear GP and propose a canonical transformation that converts original linear representations into a canonical form in which structural redundancies are removed. In canonical form, we can easily verify whether two representations represent an identical program. We then discuss exploitation of the proposed canonical transformation, and demonstrate a way to improve search performance of linear GP by avoiding redundant individuals. Experiments were conducted with an image feature synthesis problem. Firstly, we have verified that there are really a lot of redundancies in conventional linear GP. We then investigate the effect of avoiding redundant individuals. The results yield that linear GP with avoidance of redundant individuals obviously outperforms conventional linear GP.

Vehicle detection and classification system based on virtual detection zone

This paper proposes a vehicle detection and classification system based on virtual detection zone (VDZ). The proposed system consists of four main steps: foreground extraction, vehicle detection, vehicle feature extraction and vehicle classification. A moving vehicle is firstly detected based on Gaussian mixture model (GMM). Then, several techniques including region of interest selection, adaptive morphological operation, and contour processing are applied to obtain correct foreground objects. Next, vehicle features are calculated when the centroid of a vehicle is on the VDZ. Finally, vehicles are classified by using k-nearest neighbor classifier. Experimental results show that the proposed method can accurately detect and classify vehicles with an accuracy of 98.53%.

Interest Point Detection Based on Stochastically Derived Stability

We propose a novel framework called StochasticSIFT for detecting interest points (IPs) in video sequences. The proposed framework incorporates a stochastic model considering the temporal dynamics of videos into the SIFT detector to improve robustness against fluctuations inherent to video signals. Instead of detecting IPs and then removing unstable or inconsistent IP candidates, we introduce IP stability derived from a stochastic model of inherent fluctuations to detect more stable IPs. The experimental results show that the proposed IP detector outperforms the SIFT detector in terms of repeatability and matching rates.

Shape recognition by using Scale Invariant Feature Transform for contour

This paper proposes a novel shape feature extractor named Contour-SIFT along with a matching method that computes the similarity between two set of proposed descriptors. It allows a shape to be recognized based on automatically located outstanding local features on its contour, which are extracted from 1-D signal representations of different smoothing scales. The algorithm describes each local feature as a list of frequencies from curvature histogram, which is created from curve segment around each local position. The descriptors will give high similarity compared with a model descriptors of a similar shape. The algorithm has properties of image scaling-, translation-, and rotation-invariants. An experiment were conducted with 200 images from Flavia dataset for verification. The result of using the proposed algorithm is compared with the result of using CSS.

Separation of occluded leaves using direction field

This paper proposes a method that separates the region of each leaf from an image of occluded leaves and produces a set of single-leaf images as an output. To identify the region of a single leaf, intersection points and direction field are required. An intersection point, which is defined as a concave point between leaves, is used as the starting position of leaf estimation process. Direction field, which describes the average direction of edges in a local area, is used to guide the estimation process. Leaf separation process applies the result of leaf estimation process to create an output. Experimental results show that 71.23% of testing leaf images were correctly separated from each other with a segmentation accuracy of 88.80%.

Image analysis algorithms for vehicle color recognition

This work discusses about the implementation of a vehicle color recognition system to be used with the vehicle license plate recognition system. Because of the complexity of Thai alphabet, current license plate recognition system fails in correctly recognizing vehicles that have the same number but different Thai characters. Also the current system cannot identify illegal number plates that are being used. This color recognition system will help to resolve these problems and increase the accuracy of the vehicle recognition system. In addition, this system will provide a wide range of color classification which includes 13 colors including white, silver and gray colors. Color recognition is done by two methods: one with machine learning and one without machine learning. A best accuracy of 87.52% is given when using SVM to classify colors.