Earticle

최근에는 강력한 특징 추출 방법에 대한 연구가 활발히 진행되고 있다. 대다수의 최신 방법들에서는 회색조 영상에 서의 계산 비용을 줄이는 방법에 대해 다루고 있다. 하지만 이러한 변환 방법은 정보가 손실되고 매칭 성능에 영향 을 줄 수 있다. 본 연구에서는 비선형 스케일 공간에서 다중 스케일 2D 불변 컬러 검출기와 기술어를 제안한다. 본 연구의 알고리즘은 HLS (색조, 밝기 및 채도) 공간의 색상 정보를 활용한다. 비선형 스케일 공간은 각 색상 채널에 대해 개별적으로 구축되며 적응형 Hessian 응답을 키포인트 추출에 사용된다. 부분적으로 적응형 흐림 효과를 나 타내기 위해 영상 데이터에서 중요한 가장자리를 보존하면서 영상 노이즈를 줄일 수 있는 FED (Fast Explicit Diffusion) 방법을 사용한다. FED 방법을 사용함으로써 뛰어난 키포인트 지역화 정확성과 특색 있는 특징을 찾을 수 있다. 또한, 본 연구에서는 HLS 색상 정보와 비선형 스케일 공간의 기울기 정보를 결합한 CM-LMB (Color Modified-Local Difference Binary) 기술자를 제안한다. 본 연구에서 제안한 시스템 (HLS-AKAZE)은 표준 영 상 데이터셋에서 뛰어난 성능을 보여 주며, 특정 이미지 (Less-Informative 그레이 스케일 이미지)에서 성능이 향 상되었다. 또한, 본 연구에서 제안하는 방법은 회전 및 스케일 그리고 조명에 대하여 불변성을 보인다.

In recent years, there have been significant research on robust feature extraction methods. Most state-of-the-art methods operate on grayscale images to decrease the computational expenses. We observe that this conversion can cause information loss and effect the matching performance. In this study, a multi-scale 2D invariant color detector and descriptor in nonlinear scale spaces is proposed. The algorithm exploits color information in HLS (Hue, Lightness, and Saturation) space. Nonlinear scale spaces are built separately for each color channel and adaptive Hessian responses are calculated for keypoint extraction. Fast Explicit Diffusion (FED) scheme is used for locally adaptive blurring to the image data. FED reduces image noise while preserving important edges. This aids in superior keypoint localization accuracy and feature distinctiveness. In addition, a Color Modified-Local Difference Binary (CM-LMB) descriptor is proposed. It combines HLS color information with gradient information in nonlinear scale space. Our proposed system (HLS-AKAZE) shows comparable performance in standard image datasets and good performance improvements in certain images (Less-Informative Grayscale Images). HLS-AKAZE is rotation, scale, and illumination invariant.

For several years, keyboard and mouse have been used as the main interacting devices between users and computer games, but they are becoming outdated. Gesture-based human- computer interaction systems are becoming more popular owing to the emergence of virtual reality and augmented reality technologies. Therefore research on these systems has attracted a significant attention. The researches focus on designing the interactive interfaces between users and computers. Human-computer interaction is an important factor in computer games because it affects not only the experience of the users, but also the design of the entire game. In this research, we develop an particle filter-based face tracking method using color distributions as features, for the purpose of applying to gesture-based human-computer interaction systems for computer games. The experimental results proved the efficiency of particle filter and color features in face tracking, showing its potential in designing human-computer interactive games.

본 논문은 영상 속 객체와 배경의 컬러 특징을 이용한 주요 객체의 자동 추출 방법에 관한 연구이다. 인간 이 객체를 판단할 때에는 배경과 객체의 색상 차이를 이용하는데 이러한 요소를 객체 추출 방법에 적용시키기 위해 서는 배경과 객체의 색차를 강조하여야 한다. 따라서 본 논문에서는 원 RGB 영상을 인간의 시각 시스템과 유사한 HSV 색 공간으로 변환하고 각기 다른 분포도의 메디안 필터를 적용한 두 개의 영상을 생성한 뒤 두 개의 메디안 필 터가 적용된 영상들을 합산하였고 데이터 군집화 방법인 Mean Shift 알고리즘을 적용하여 색상 특징을 그룹화 하였 다. 마지막으로 이진화 작업을 위하여 영상의 채널 수를 3 채널에서 1 채널로 정규화 한 뒤 영상 내 픽셀들의 평균 값을 임계값으로 이용하는 이진화 방법으로 객체 지도 영상을 생성하였고 주요 객체를 추출하였다.

This paper is a study on an object extraction method which using color features of an object and background in the image. A human recognizes an object through the color difference of object and background in the image. So we must to emphasize the color's difference that apply to extraction result in this image. Therefore, we have converted to HSV color images which similar to human visual system from original RGB images, and have created two each other images that applied Median Filter and we merged two Median filtered images. And we have applied the Mean Shift algorithm which a data clustering method for clustering color features. Finally, we have normalized 3 image channels to 1 image channel for binarization process. And we have created object map through the binarization which using average value of whole pixels as a threshold. Then, have extracted major object from original image use that object map.

본 논문에서는 컬러정보에 기반한 분포맵과 얼굴 구성요소 인식을 위한 평가함수를 이용한 얼굴 검출 방법을 제안 한다. 제안한 방법은 먼저 조명보정 기법과 피부색 분포맵을 이용하여 얼굴 후보영역을 검출하고 검출된 얼굴 후보 영역에 생긴 홀 영역을 이용하여 눈 후보영역을 검출한다. 눈 후보영역에서 눈 영역과 눈 주위 영역을 이용한 템플 릿 정합 방법으로 눈 후보점을 검출하고 입 영역과 입 주위 영역의 색상 분포에 기반한 입 인식 평가함수를 이용하 여 입을 검출하여 최종적으로 얼굴 영역을 검출한다. 제안한 방법의 유효성을 검증하기 위하여 웹에서 획득한 영상 을 이용하여 실험한 결과 좋은 성능을 보였다.

In this paper, a new face detection method using distribution maps based on color information and evaluation function for face features is proposed. Face-like regions are first extracted by using lighting compensation technique and skin color distribution maps and then, eye-like regions are extracted by using hole regions generated within the face-like regions. Eye candidates are then detected by means of a template matching method based on eye regions and rectangular regions around eye regions within eye-like regions. Finally, face regions are detected by using evaluation values of two eye candidates and evaluation function for mouth recognition based on color distributions of mouth region and region around mouth. Experimental results show that the proposed method can achieve a high performance.

The computer vision techniques are required for the identification of leafy vegetables for the development of veggie vision applications. This paper presents the mean and range color features based identification of leafy vegetables. Initially, a total of 18 RGB and HSI color features are chosen. A reverse engineering process is adopted for reduction of features. Finally 12 mean and range features of RGB and HSI color features are selected based on the performance. A BPNN based classifier is used for identification of vegetables. The identification rate is in the range 92-100% for ten types of vegetables. The work finds applications in automatic vending, packing and grading of vegetables, food preparation and the like.

Construct an image retrieval method based on feature fusion of color features of RGB component and edge feature. The execution of this method is divided into the following steps: First, the original image is decomposed into three components images of RGB; Secondly, wavelet decomposition is carried out on each component image; Third, color and edge features are obtained on the low-pass components from each component image; Finally, these two features are combined to complete the image retrieval. Experimental results show that the proposed method has better accuracy than the retrieval method using single feature, and retrieval speed is relatively fast.

The analysis of MRI images is a manual process carried by experts which need to be automated to accurately classify the normal and abnormal images. We have proposed a reduced, three staged model having pre-processing, feature extraction and classification steps. In preprocessing the noise has been removed from grayscale images using a median filter, and then grayscale images have been converted to color (RGB) images. In feature extraction, red, green and blue channels from each channel of the RGB has been extracted because they are so much informative and easier to process. The first three color moments mean, variance, and skewness are calculated for each red, green and blue channel of images. The features extracted in the feature extraction stage are classified into normal and abnormal with K-Nearest Neighbors (k-NN). This method is applied to 100 images (70 normal, 30 abnormal). The proposed method gives 98.00% training and 95.00% test accuracy with datasets of normal images and 100% training and 90.00% test accuracy with abnormal images. The average computation time for each image was .06s.

Many textures based image retrieval researchers use global texture features for representing and retrieval of images from an image database. However, this leads to misrepresentation of local information leading to the inefficient image retrieval performance. This paper presents an approach to overcome the problem. The approach focuses on extracting local Haralick’s texture feature based on a predetermined region using the color co-occurrence matrix method, the selection of the ‘significant’ Haralik’s texture features and evaluation of the performance of the combination of the ‘significant’ features. The proposed method which is an Average Analysis and a well known method, Principal Component Analysis were applied to obtain ‘significant’ features. In order to compare the performance, a series of experiments were carried out for both methods, which is the proposed Average Analysis and the Principal Component Analysis. Experiments were performed on a 1000 selected images from the Coral image database which were divided into ten categories. Based on the experimental results, it is interesting to note that for the combination ‘significant’ features obtained from the proposed Average Analysis showed better retrieval performance compared to the Principal Component Analysis for almost all categories. This finding has an important implication in deciding the correct combination of ‘significant’ features for certain image properties. It has shown that the proposed method is able to produce less computational processing time due to a reduced amount of processing involved. The result is also compared to the previous researches and has shown an increase of an average precision from 8.5% to 26%.

This paper describes Support Vector Machine (SVM) and Artificial Neural Network (ANN) based recognition and classification of visual symptoms affected by fungal disease. Color images of fungal disease symptoms affected on cereals like wheat, maize and jowar are used in this work. Different types of symptoms affected by fungal disease namely leaf blight, leaf spot, powdery mildew, leaf rust, smut are considered for the study. The developed algorithms are used to preprocess, segment, extract features from disease affected regions. The affected regions are segmented using k-means segmentation technique. Color texture features are extracted from affected regions and then used as inputs to SVM and ANN classifiers. The texture analysis is done using Color Co-occurrence Matrix. Tests are performed to classify image samples. Classification accuracies between 68.5% and 87% are obtained using ANN classifier. The average classification accuracies have increased to 77.5% and 91.16% using SVM classifier.

This paper presents a methodology to recognize certain crop fields’ images using texture, color and combination of both types of features. In this work, we have considered eight varieties of crop images, namely, Brinjal, Cotton, Groundnut, Paddy, Soyabean, Sugarcane and Sunflower. Texture features using GLCM and color features using HSV are deployed. Artificial Neural Network (ANN) is used for recognition. Considering only as feature, classification accuracies of 63.75%, 66.25% and 84.375% are obtained using texture, color and their combination respectively. The work is helpful in the area of agriculture for early detection and prevention of diseases.

Computer vision applications in the field of agriculture science are gaining importance. The paper presents a method for recognition of paddy varieties from bulk paddy grain image samples based on color texture features extracted from color co-occurrence matrices. The color texture features are obtained from H, S and I color planes and their combinations. The feature set is reduced based on contribution of features to the recognition accuracy. The reduced feature set of the HS plane includes Energy, Entropy and Correlation features from Hue plane and Energy, Entropy, Contrast, and Correlation features from Saturation plane. The paddy grain images are recognized using a multilayer feed-forward artificial neural network. The considered fifteen paddy varieties have given the recognition accuracy of 92.33%. The work is useful in developing a machine vision system for agriculture produce market and developing multimedia applications in agriculture sciences.

Based on the problems that target appears rotation and noise interference in complex environment, an improved multi-feature adaptive fusion tracking method is proposed. The algorithm adopts unscented Kalman particle filter (UPF) to update the measurement information in the sample particles, better overcome the problem of the particle weight degradation. In addition, in order to overcome the defects of additive and multiplicative fusion algorithm in the feature selection, the multiple adaptive fusion characteristics method that target color distribution and scale invariance feature (SIFT) are used as complementary information. Experimental results show that the proposed method is superior to the traditional methods which are based on fixed weight or standard particle filter.

A novel approach of Content Based Image Retrieval(CBIR), which combines color, texture and shape descriptors to represent the features of the image, is discussed in this paper. The proposed scheme is based on three noticeable algorithms: color distribution entropy(CDE), color level co-occurrence(CLCM) and invariant moments. CDE takes the correlation of the color spatial distribution in an image into consideration. CLCM matrix is the texture feature of the image, which is a new proposed descriptor that is grounded on co-occurrence matrix to seize the alteration of the texture. Hu invariant moments are frequently used owing to its invariance under translation, changes in scale, and also rotation. The proposed scheme achieves a modest retrieval result by utilizing these diverse and primitive image descriptors, at the same time, the retrieval result is better when use the texture feature alone which we proposed than use gray level co-occurrence. The similarity measure matrix is based upon Euclidean distance.

For pose-varied color face image, this paper proposed a method of facial feature location and pose estimation based on an unsupervised sphere skin model and fused facial features. Firstly, an adaptive preprocessing, an established unsupervised sphere model and the holes filling technology were presented to extract the face. Then the difference feature and different local binary pattern textural feature of human face were fused to used in the hybrid integral projection technology for localizing the facial features. Finally, poses were estimated using the geometrical distribution of facial features. Experimental results show that, the method can locate facial features and classify different pose more effectively and adaptively.

기존 논문은 컬러 채널 변환에 의한 상관관계 밀접도로 영상을 검색 하였다. 기존의 논문의 문제점은 공간적 분포가 반영이 되지 않고, 컬러 색상 상관관계 분석으로 상관도가 높으면 같은 영상으로 검색하는 문제점이 있다. 본 논문에서는 쿼드 트리와 컬러 특징을 이용한 영상 검색 방법을 제안하고자 한다. 컬러 특징 정보의 상관관계를 분석하여 1차 검색 영상을 선택한 후, 쿼드 트리의 공간 정보의 상호 유사성을 비교하여 2차 검색함으로서 컬러의 특성과 공간적 특성을 반영하는 검색 방법을 제안하고자 한다.

Conventional methods for image retrieval use the proximity of correlation feature by transforming color channel. The problems in conventional methods do not reflect the features of spatial distribution and retrieve the images only if the proximity of color correlation feature is high. In this paper, we propose a retrieval method using quad-tree and color features. In the first step, we retrieval images using color correlation feature. In the second step, we finally retrieve the images among selected images in the first step using similarity of quad-tree.

In this paper, we propose a detection method for facial features in color images with various backgrounds and face poses. To begin with, the proposed method extracts face candidacy region from images with various backgrounds, which have skin-tone color and complex objects, via the color and edge information of face. And then, by using the elliptical shape property of face, we correct a rotation, scale, and tilt of face region caused by various poses of head. Finally, we verify the face using features of face and detect facial features. In our experimental results, it is shown that accuracy of detection is high and the proposed method can be used in pose-invariant face recognition system effectively

It has been suggested that merging plays an important role in the formation and the evolution of early-type galaxies (ETGs). Optical-NIR color gradients of ETGs in high density environments are found to be less steep than those of ETGs in low density environments, hinting frequent merger activities in ETGs in high density environments. In order to examine if the flat color gradients are the result of dry mergers, we studied the relations between merging features, luminosities, environments and color gradients of 196 low redshift ETGs selected from Sloan Digital Sky Survey (SDSS) Stripe82. Near Infrared (NIR) images are taken from UKIRT Infrared Deep Sky Survey (UKIDSS) Large Area Survey (LAS). Color (r-K) gradients of ETGs with tidal features are a little flatter than relaxed ETGs, but not significant. We found that massive (> 10^11.3 solar masses) ETGs have -40% less scattered color gradients than less massive ETGs. The less scattered color gradients of massive ETGs could be evidence of dry merger processes in the evolution of massive ETGs. We found no relation between color gradients of ETGs and their environments.

It has been suggested that merging plays an important role in the formation and the evolution of early-type galaxies (ETGs). Optical-NIR color gradients of ETGs in high density environments are found to be less steep than those of ETGs in low density environments, hinting frequent merger activities in ETGs in high density environments. In order to examine if the flat color gradients are the result of dry mergers, we studied the relations between merging features, color gradient, and environments of 281 low redshift ETGs selected from Sloan Digital Sky Survey (SDSS) Stripe82. The sample contains 222 relaxed ETGs, 38 ETGs with tidal features, 10 galaxies with dust features and 11 galaxies with tidal and dust features, and Near Infrared (NIR) images are taken from UKIRT Infrared Deep Sky Survey (UKIDSS) Large Area Survey (LAS). We find that r-K color gradients of field sample galaxies are steeper than those of sample ETGs within cluster environments. For the field sample galaxies, a relatively large number of galaxies with peculiar features contribute to the steeper color gradients, while the absence of these peculiar early-type galaxies make color gradients of the cluster sample galaxies intact. In high density environment, ETGs are already evolved and relaxed, resulting flat color gradients. However, in low density environments, a majority of ETGs undergone merging recently which makes the color gradients steep.

It has been suggested that merging plays an important role in the formation and the evolution of early-type galaxies. Optical-NIR color gradients of early-type galaxies in high density environments are found to be less steep than those in low density environment, hinting frequent merger activities in early-type galaxies in high density environment. In order to confirm if the flat color gradient is the result of dry merger, we decided to look deeply to find merging features and get their relation with color gradient. We selected samples which show extreme values of optical-NIR color gradients based on the data of previous study, and observed them at Maidanak observatory 1.5m telescope with long exposure. After masking out overlaid sources, our analysis reveals that these galaxies do not have extreme color gradient values. High degree sky flat technique was used during observation to aid discovery of faint, extended features. However, flatness of detector (SNUCAM) was good enough, so we could not see any marked improvement in image quality compared to those using normal sky flats. Additionally we noticed a feature that looks like merging tidal tail in the CFHT archival image, but this does not show up on the image we obtained. This demonstrates that flatness and correct sky estimation is very important when we look for faint merging features. In future we plan to enlarge the number of the sample.

In this paper, we present a new texture image retrieval method which combines color and texture features extracted from images by a set of multi-resolution multi-direction (MRMD) filters. The MRMD filter set chosen is simple and can be separable to low and high frequency information, and provides efficient multi-resolution and multi-direction analysis. The color space used is HSV color space separable to hue, saturation, and value components, which are easily analyzed as showing characteristics similar to the human visual system. This experiment is conducted by comparing precision vs. recall of retrieval and feature vector dimensions. Images for experiments include Corel DB and VisTex DB; Corel_MR DB and VisTex_MR DB, which are transformed from the aforementioned two DBs to have multi-resolution images; and Corel_MD DB and VisTex_MD DB, transformed from the two DBs to have multi-direction images. According to the experimental results, the proposed method improves upon the existing methods in aspects of precision and recall of retrieval, and also reduces feature vector dimensions.