• Journal of KIISE
• /
• v.42 no.2
• /
• pp.203-212
• /
• 2015

This paper proposes a traffic sign recognition method in real road environments. The video stream in driving environments has two different characteristics compared to a general object video stream. First, the number of traffic sign types is limited and their shapes are mostly simple. Second, the camera cannot take clear pictures in the road scenes since there are many illumination changes and weather conditions are continuously changing. In this paper, we improve a modified census transform(MCT) to extract features effectively from the road scenes that have many illumination changes. The extracted features are collected by histograms and are transformed by the dense descriptors into very high dimensional vectors. Then, the high dimensional descriptors are encoded into a low dimensional feature vector by Fisher-vector coding and Gaussian Mixture Model. The proposed method shows illumination invariant detection and recognition, and the performance is sufficient to detect and recognize traffic signs in real-time with high accuracy.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.2
• /
• pp.33-44
• /
• 2010

This paper propose the structure of real-time face detection hardware architecture for robot vision processing applications. The proposed architecture is robust against illumination changes and operates at no less than 60 frames per second. It uses Modified Census Transform to obtain face characteristics robust against illumination changes. And the AdaBoost algorithm is adopted to learn and generate the characteristics of the face data, and finally detected the face using this data. This paper describes the face detection hardware structure composed of Memory Interface, Image Scaler, MCT Generator, Candidate Detector, Confidence Comparator, Position Resizer, Data Grouper, and Detected Result Display, and verification Result of Hardware Implementation with using Virtex5 LX330 FPGA of Xilinx. Verification result with using the images from a camera showed that maximum 32 faces per one frame can be detected at the speed of maximum 149 frame per second.

• Journal of IKEEE
• /
• v.26 no.4
• /
• pp.686-692
• /
• 2022

In this paper, the matching error rates of modified area-based, energy-based algorithms, and learning-based structures were compared for stereo image matching. Census transform (CT) based on region and life propagation (BP) algorithm based on energy were selected, respectively.Existing algorithms have been improved and implemented in an embedded processor environment so that they can be used for stereo image matching in mobile systems. Even in the case of the learning base to be compared, a neural network structure that utilizes small-scale parameters was adopted. To compare the error rates of the three matching methods, Middlebury's Tsukuba was selected as a test image and subdivided into non-occlusion, discontinuous, and disparity error rates for accurate comparison. As a result of the experiment, the error rate of modified CT matching improved by about 11% when compared with the existing algorithm. BP matching was about 87% better than conventional CT in the error rate. Compared to the learning base using neural networks, BP matching was about 31% superior.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.48 no.4
• /
• pp.116-128
• /
• 2011

In this paper, we propose the structure of a high-performance face-detection engine that responds well to facial rotating changes using rotation transformation which minimize the required memory usage compared to the previous face-detection engine. The validity of the proposed structure has been verified through the implementation of FPGA. For high performance face detection, the MCT (Modified Census Transform) method, which is robust against lighting change, was used. The Adaboost learning algorithm was used for creating optimized learning data. And the rotation transformation method was added to maintain effectiveness against face rotating changes. The proposed hardware structure was composed of Color Space Converter, Noise Filter, Memory Controller Interface, Image Rotator, Image Scaler, MCT(Modified Census Transform), Candidate Detector / Confidence Mapper, Position Resizer, Data Grouper, Overlay Processor / Color Overlay Processor. The face detection engine was tested using a Virtex5 LX330 FPGA board, a QVGA grade CMOS camera, and an LCD Display. It was verified that the engine demonstrated excellent performance in diverse real life environments and in a face detection standard database. As a result, a high performance real time face detection engine that can conduct real time processing at speeds of at least 60 frames per second, which is effective against lighting changes and face rotating changes and can detect 32 faces in diverse sizes simultaneously, was developed.

• Annual Conference of KIPS
• /
• 2011.11a
• /
• pp.322-325
• /
• 2011

스테레오 비전 시스템에서 전통적인 매칭 알고리즘으로 SAD(Sum of Absolute Differences), SSD(Sum of Squared Differences), NCC(Normalized Cross Correlation) 등 다양한 알고리즘이 존재한다. 그러나 하드웨어로 실시간 처리를 위한 시스템을 구현하기 위해서는 리소스가 한정 되어있다는 제약 때문에 많은 연구에서 SAD 혹은 RT(Rank Transform), CT(Census Transform)를 많이 사용하게 된다. FPGA 내부에는 BRAM(Block RAM)과 MAC(multiply-accumulator)인 DSP슬라이스가 이미 존재한다. 본 논문에서는 BRAM과 DSP로직을 활용해서 전통적인 매칭 알고리즘 중에서 연산기 사용이 가장 많은 NCC를 FPGA로 실시간 처리 가능한 하드웨어 구조를 제안한다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.25 no.1
• /
• pp.91-96
• /
• 2015

In this study, we introduce robust face recognition system with illumination variation realized with the aid of CT preprocessing method. As preprocessing algorithm, Census Transform(CT) algorithm is used to extract locally facial features under unilluminated condition. The dimension reduction of the preprocessed data is carried out by using $(2D)^2$PCA which is the extended type of PCA. Feature data extracted through dimension algorithm is used as the inputs of proposed radial basis function neural networks. The hidden layer of the radial basis function neural networks(RBFNN) is built up by fuzzy c-means(FCM) clustering algorithm and the connection weights of the networks are described as the coefficients of linear polynomial function. The essential design parameters (including the number of inputs and fuzzification coefficient) of the proposed networks are optimized by means of artificial bee colony(ABC) algorithm. This study is experimented with both Yale Face database B and CMU PIE database to evaluate the performance of the proposed system.

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.12
• /
• pp.71-82
• /
• 2012

In this paper, we propose a license plate detection method with improved Adaboost learning and MCT (Modified Census Transform). The MCT represents the local structure patterns as integer numbered feature values which has robustness to illumination change and memory efficiency. However, since these integer values are discrete, a lookup table is needed to design a weak classifier for Adaboost learning. Some previous research efforts have focused on minimization of exponential criterion for Adaboost optimization. In this paper, a method that uses MCT and improved Adaboost learning based on Newton's optimization to exponential criterion is proposed for license plate detection. Experimental results on license patch images and field images demonstrate that the proposed method yields higher performance of detection rates with low false positives than the conventional method using the original Adaboost learning.

• Journal of Korea Multimedia Society
• /
• v.16 no.3
• /
• pp.302-308
• /
• 2013

In this research, we have explored personal authentication system through multimodal biometrics for mobile computing environment. We have selected face and speaker recognition for the implementation of multimodal biometrics system. For face recognition part, we detect the face with Modified Census Transform (MCT). Detected face is pre-processed through eye detection module based on k-means algorithm. Then we recognize the face with Principal Component Analysis (PCA) algorithm. For speaker recognition part, we extract features using the end-point of voice and the Mel Frequency Cepstral Coefficient (MFCC). Then we verify the speaker through Dynamic Time Warping (DTW) algorithm. Our proposed multimodal biometrics system shows improved verification rate through combining two different biometrics described above. We implement our proposed system based on Android environment using Galaxy S hoppin. Proposed system presents reduced false acceptance ratio (FAR) of 1.8% which shows improvement from single biometrics system using the face and the voice (presents 4.6% and 6.7% respectively).

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.7
• /
• pp.67-77
• /
• 2016

In this paper, we propose an algorithm for open/closed eye detection based on modified Hausdorff distance. The proposed algorithm consists of two parts, face detection and open/closed eye detection parts. To detect faces in an image, MCT (Modified Census Transform) is employed based on characteristics of the local structure which uses relative pixel values in the area with fixed size. Then, the coordinates of eyes are found and open/closed eyes are detected using MHD (Modified Hausdorff Distance) in the detected face region. Firstly, face detection process creates an MCT image in terms of various face images and extract criteria features by PCA(Principle Component Analysis) on offline. After extraction of criteria features, it detects a face region via the process which compares features newly extracted from the input face image and criteria features by using Euclidean distance. Afterward, the process finds out the coordinates of eyes and detects open/closed eye using template matching based on MHD in each eye region. In performance evaluation, the proposed algorithm achieved 94.04% accuracy in average for open/closed eye detection in terms of test video sequences of gray scale with 30FPS/$320{\times}180$ resolution.

• ETRI Journal
• /
• v.38 no.2
• /
• pp.347-355
• /
• 2016

In face examinations, gender classification (GC) is one of several fundamental tasks. Recent literature on GC primarily utilizes datasets containing high-resolution images of faces captured in uncontrolled real-world settings. In contrast, there have been few efforts that focus on utilizing low-resolution images of faces in GC. We propose a GC method based on a pixel classifier boosting with modified census transform features. Experiments are conducted using large datasets, such as Labeled Faces in the Wild and The Images of Groups, and standard protocols of GC communities. Experimental results show that, despite using low-resolution facial images that have a 15-pixel inter-ocular distance, the proposed method records a higher classification rate compared to current state-of-the-art GC algorithms.