• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2003년도 하계종합학술대회 논문집 Ⅳ
• /
• pp.2307-2310
• /
• 2003

In this paper, a method for face recognition based on the wavelet packet decomposition is proposed. In the proposed method, the input image is decomposed by the 2-level wavelet packet transformation and then the face areas are defined by the Integral Projection technique applied to each of the 1-level subband images, HL and LH. After the defined face areas are divided into three areas, called top, bottom, and border, the mean and the variance of the three areas of the approximation image are computed, and the variance of the single predetermined face area for the rest of 15 detail images, from which the feature vectors of statistical measure are extracted. In this paper we use the wavelet packet decomposition, a generalization of the classical wavelet decomposition, to obtain its richer signal analysis features such as discontinuity in higher derivatives, self-similarity, etc. And we have shown that even with very simple statistical features such as mean values and variance we can make an excellent basis for face classification, if an appropriate probability distance is used.

• 한국농업기계학회:학술대회논문집
• /
• 한국농업기계학회 1996년도 International Conference on Agricultural Machinery Engineering Proceedings
• /
• pp.852-862
• /
• 1996

This study was carried out to develop tools to detect defects of apple using machine vision. For the purpose, 6 kinds of frame for color images, R, G, B, h, S, and I frame, and a frame for near infra-red images (NIR frame) were tested first to select one which is useful to segment defect areas from apple images. After then, several methods to classify kind of defect for the segmented defect areas were developed and tested. Five kinds of apple defect -bruise , decay ,fleck worm hole and scar were investigated . The results are as follows: NIR frame was selected as the best one among the 7 kinds of image frame, and R, G and I frames showed favourable result to segment areas of apple defect. Various features of the segmented defect areas were measured to classify the defect areas. Eight kids of feature of the areas-size, roundness, axes length ratio, mean and variance of pixel values, variance of real part of spectrum, mean and variance of power spectrum resulted from spacial ourier transform were observed for the segmented defect areas in the selected 4 frames. then procedures to classify defects using the features were developed for the 4 frames and tested with 75-113 defects on apples. The test resulted that NIR and I frames showed high accuracies to classify the kind of defect as 77% and 76% , respectively.

• 대한임베디드공학회논문지
• /
• 제3권1호
• /
• pp.8-18
• /
• 2008

In this paper, we propose an efficient algorithm using wavelet transform for still image database retrieval. Especially, it uses only the lowest frequency sub-band in multi-level wavelet transform so that a retrieval system uses a smaller quantity of memory and takes a faster processing time. We extract different textured features, statistical information such as mean, variance and histogram, from low frequency sub-band. Then we measure the distances between the query image and the images in a database in terms of these features. To obtain good retrieval performance, we use the first feature (mean and variance of wavelet coefficients) to filter out most of the unlikely images. The rest of the images are considered to be candidate images. Then we apply the second feature (histogram of wavelet coefficient) to rank all the candidate images. To evaluate the algorithm, we create various distorted image databases using MIT VisTex texture images and PICS natural images. Through simulations, we demonstrate that our method can achieve performance satisfactorily in terms of the retrieval accuracy as well as the both memory requirement and computational complexity. Therefore it is expected to provide good retrieval solution for JPEG-2000 using wavelet transform.

• 한국통신학회논문지
• /
• 제35권7C호
• /
• pp.604-610
• /
• 2010

본 논문에서는 기존의 CLSFN (Cepstral distance and Log-energy based Silence Feature Normalization) 방법의 인식성능을 향상시키기 위하여, 필터 뱅크 서브 밴드 영역에서 잡음을 차감하는 방법과 CLSFN을 결합하는 방법, 즉 FSFN (Filter bank sub-band energy subtraction based CLSFN)을 제안하였다. 이 방법은 음성으로부터 특징 파라미터를 추출할 때 필터 뱅크 서브 밴드 영역에서 잡음을 제거하여 켑스트럼 특징을 향상시키고, 이에 대한 켑스트럼 거리를 이용하여 음성/묵음 분류의 정확도를 개선함으로써 기존 CLSFN 방법에 비해 향상된 인식성능을 얻을 수 있다. Aurora 2.0 DB를 이용한 실험결과, 제안하는 FSFN 방법은 CLSFN 방법에 비해 평균 단어 정확도 (word accuracy)가 약 2% 향상되었으며, CMVN (Cepstral Mean and Variance Normalization)과의 결합에서도 기존 모든 방법에 비해 가장 우수한 인식성능을 나타내어 제안 방법의 유효성을 확인할 수 있었다.

• 한국산학기술학회논문지
• /
• 제16권5호
• /
• pp.3378-3384
• /
• 2015

최근에 라만스펙트럼에 대한 고속 검색 방법은 많은 관심을 받아왔다. 지금까지 가장 간단하고 널리 사용되는 방법은 주어진 스펙트럼과 데이터베이스 스펙트라 사이의 유클리드 거리를 계산하고 비교하는 방법이다. 하지만 고차원 데이터의 속성으로 검색의 문제는 그리 간단하지 않다. 가장 큰 문제점중의 하나는 검색 방법에 있어서 연산량이 많아 계산 시간이 너무 오래 걸린다는 것이다. 이러한 문제점을 극복하기 위해, 코드워드의 MPS(Mean Pyramids Search)와 PDS(Partial Distortion Search)을 사용하는 알고리즘이 현재 이미지 코딩 분야에서 고속 검색 알고리즘으로 널리 사용되고 있다. 하지만 이 방법은 1차원 데이터의 경우에는 적합하지 않다. 본 논문에서 우리는 라만 스펙트럼 데이터에 적합한 3가지 새로운 방법의 고속 검색 알고리즘을 제안한다. 이 방법은 벡터의 두 개의 주요한 특징으로 평균과 분산을 사용하여 후보가 될 수 없는 많은 코드워드를 계산하지 않으므로 연산량을 줄이고 계산 시간을 줄여준다. 실험은 1DMPS+PDS와 비교하여 1DMPS Sort+PDS는 42.8%, 1DMPS Sort+PDS는 48.6%, 1DMPS Sort with Sorted Variance+PDS는 55.2%의 성능향상을 보였다. 실험결과는 제안된 알고리즘이 고속 검색에 적합함을 확인시켜 준다.

• International Journal of Contents
• /
• 제4권2호
• /
• pp.24-28
• /
• 2008

In this paper, we propose a new approach for content-based video retrieval using non-parametric based motion classification in the shot-based video indexing structure. Our system proposed in this paper has supported the real-time video retrieval using spatio-temporal feature comparison by measuring the similarity between visual features and between motion features, respectively, after extracting representative frame and non-parametric motion information from shot-based video clips segmented by scene change detection method. The extraction of non-parametric based motion features, after the normalized motion vectors are created from an MPEG-compressed stream, is effectively fulfilled by discretizing each normalized motion vector into various angle bins, and by considering the mean, variance, and direction of motion vectors in these bins. To obtain visual feature in representative frame, we use the edge-based spatial descriptor. Experimental results show that our approach is superior to conventional methods with regard to the performance for video indexing and retrieval.

• 펄프종이기술
• /
• 제32권5호
• /
• pp.14-25
• /
• 2000

In this investigation, a wavelet transform analysis was used to decompose beta-radiographic formation images into spectral and spatial components. Conventional formation analysis may use spectral analysis, based on Fourier transformation or variance vs. zone size, to describe the grammage distribution of features such as flocs, streaks and mean fiber orientation. However, these methods have limited utility for the analysis of statistically stationary data sets where variance is not uniform with position, e.g. paper machine CD profiles (especially those that contain streaks). A continuous wavelet transform was used to analyze formation data arrays obtained from radiographic imaging of handsheets and cross machine paper samples. The response of the analytical method to grammage, floc size distribution, mean fiber orientation an sensitivity to feature localization were assessed. From wavelet analysis, the change in scale of grammage variation as a function of position was used to demonstrate regular and isolated differences in the formed structure.

• 농업과학연구
• /
• 제49권4호
• /
• pp.785-793
• /
• 2022

In this study, tomato maturity was estimated based on deep learning for a harvesting robot. Tomato images were obtained using a RGB camera installed on a monitoring robot, which was developed previously, and the samples were cropped to 128 × 128 size images to generate a dataset for training the classification model. The classification model was constructed based on convolutional neural networks, and the mean-variance loss was used to learn implicitly the distribution of the data features by class. In the test stage, the tomato maturity was estimated as a continuous index, which has a range of 0 to 1, by calculating the expected class value. The results show that the F1-score of the classification was approximately 0.94, and the performance was similar to that of a deep learning-based classification task in the agriculture field. In addition, it was possible to estimate the distribution in each maturity stage. From the results, it was found that our approach can not only classify the discrete maturation stages of the tomatoes but also can estimate the continuous maturity.

• 대한원격탐사학회지
• /
• 제24권3호
• /
• pp.267-272
• /
• 2008

Tree species-specific estimates with spacebome high-resolution imagery improve estimation of forest biomass which is needed to predict the long term planning for the sustainable forest management(SFM). This paper is a contribution to develop crown distinguishing coniferous species, Pinus densiflora and Pinus koraiensis, from QuickBird imagery. The proposed feature analysis derived from shape parameters and first and second-order statistical texture features of the same test area were compared for the two species separation and delineation. As expected, initial studies have shown that both formfactor and compactness shape parameters provided the successful differentiating method between the pine species within the compartment for single crown identification from spaceborne high resolution imagery. Another result revealed that the selected texture parameters - the mean, variance, angular second moment(ASM) - in the infrared band image could produce good subset combination of texture features for representing detailed tree crown outline.

• 스마트미디어저널
• /
• 제8권4호
• /
• pp.64-71
• /
• 2019

The performance of an OMR (Optical Music Recognition) system is usually determined by the characterizing features of the input music score images. Low resolution is one of the main factors leading to degraded image quality. In this paper, we handle the low-resolution problem using the super-resolution technique. We propose the use of a deep neural network with instance normalization to improve the quality of music score images. We apply instance normalization which has proven to be beneficial in single image enhancement. It works better than batch normalization, which shows the effectiveness of shifting the mean and variance of deep features at the instance level. The proposed method provides an end-to-end mapping technique between the high and low-resolution images respectively. New images are then created, in which the resolution is four times higher than the resolution of the original images. Our model has been evaluated with the dataset "DeepScores" and shows that it outperforms other existing methods.