• Proceedings of the IEEK Conference
• /
• 2001.09a
• /
• pp.183-186
• /
• 2001

In this paper, we first propose new texture features, BVLC (block variation of local correlation coefficients) moments, for content-based image retrieval (CBIR) and then present an image retrieval method based on the fusion of BDIP and BVLC moments. BDIP uses the local probabilities in image blocks to extract valley and edges well. BVLC uses the variations of local correlation coefficients in images blocks to measure texture smoothness well. In order not to be affected with the movement, rotation, and size of an object, the first and second moments of BDIP and BVLC are used for CBIR. Corel DB and Vistex DB are used to evaluate the performance of the proposed retrieval method. Experimental results show that the presented retrieval method yields average 12% better performance than the method using only BDIP or BVLC moments and average 13% better performance than the method using wavelet moments.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.48 no.4
• /
• pp.76-85
• /
• 2011

In this paper, we propose a texture feature-based language identification using Gabor feature and wavelet-domain BDIP (block difference of inverse probabilities) and BVLC (block variance of local correlation coefficients) features. In the proposed method, Gabor and wavelet transforms are first applied to a test image. The wavelet subbands are next denoised by Donoho's soft-thresholding. The magnitude operator is then applied to the Gabor image and the BDIP and BVLC operators to the wavelet subbands. Moments for Gabor magnitude image and each subband of BDIP and BVLC are computed and fused into a feature vector. In classification, the WPCA (whitened principal component analysis) classifier, which is usually adopted in the face identification, searches the training feature vector most similar to the test feature vector. Experimental results show that the proposed method yields excellent language identification with rather low feature dimension for a document image DB.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.5 s.305
• /
• pp.103-114
• /
• 2005

This paper presents a content-based image retrieval (CBIR) method using the combination of color and texture features. As a color feature, a color autocorrelogram is chosen which is extracted from the hue and saturation components of a color image. As a texture feature, BDIP(block difference of inverse probabilities) and BVLC(block variation of local correlation coefficients) are chosen which are extracted from the value component. When the features are extracted, the color autocorrelogram and the BVLC are simplified in consideration of their calculation complexity. After the feature extraction, vector components of these features are efficiently quantized in consideration of their storage space. Experiments for Corel and VisTex DBs show that the proposed retrieval method yields 9.5% maximum precision gain over the method using only the color autucorrelogram and 4.0% over the BDIP-BVLC. Also, the proposed method yields 12.6%, 14.6%, and 27.9% maximum precision gains over the methods using wavelet moments, CSD, and color histogram, respectively.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.49 no.2
• /
• pp.102-112
• /
• 2012

In this paper, we propose a texture classification using wavelet-domain BDIP (block difference of inverse probabilities) and BVLC (block variance of local correlation coefficients) features with WPCA (whitened principal component analysis) classifier. In the proposed method, the wavelet transform is first applied to a query image. The BDIP and BVLC operators are next applied to the wavelet subbands. Global moments for each subband of BDIP and BVLC are then computed and fused into a feature vector. In classification, the WPCA classifier, which is usually adopted in the face identification, searches the training feature vector most similar to the query feature vector. Experimental results show that the proposed method yields excellent texture classification with low feature dimension for test texture image DBs.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.48 no.3
• /
• pp.19-29
• /
• 2011

Until now, many face recognition methods have been proposed, most of them use a 1-dimensional feature vector which is vectorized the input image without feature extraction process or input image itself is used as a feature matrix. It is known that the face recognition methods using raw image yield deteriorated performance in databases whose have severe illumination changes. In this paper, we propose a face recognition method using local statistics of gradients and correlations which are good for illumination changes. BDIP (block difference of inverse probabilities) is chosen as a local statistics of gradients and two types of BVLC (block variation of local correlation coefficients) is chosen as local statistics of correlations. When a input image enters the system, it extracts the BDIP, BVLC1 and BVLC2 feature images, fuses them, obtaining feature matrix by $(2D)^2$ PCA transformation, and classifies it with training feature matrix by nearest classifier. From experiment results of four face databases, FERET, Weizmann, Yale B, Yale, we can see that the proposed method is more reliable than other six methods in lighting and facial expression.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.9C
• /
• pp.930-938
• /
• 2005

We propose a content-based image retrieval(CBIR) method based on an efncient combination of a color feature and multiresolution texture features. As a color feature, a HSV autocorrelograrn is chosen which is blown to measure spatial correlation of colors well. As texture features, BDIP and BVLC moments are chosen which is hewn to measure local intensity variations well and measure local texture smoothness well, respectively. The texture features are obtained in a wavelet pyramid of the luminance component of a color image. The extracted features are combined for efficient similarity computation by the normalization depending on their dimensions and standard deviation vectors. Experimental results show that the proposed method yielded average $8\%\;and\;11\%$ better performance in precision vs. recall than the method using BDIPBVLC moments and the method using color autocorrelograrn, respectively and yielded at least $10\%$ better performance than the methods using wavelet moments, CSD, color histogram. Specially, the proposed method shows an excellent performance over the other methods in image DBs contained images of various resolutions.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.1 s.307
• /
• pp.55-66
• /
• 2006

We propose a progressive image retrieval method based on an efficient combination of multiresolution color and torture features in wavelet transform domain. As a color feature, color autocorrelogram of the hue and saturation components is chosen. As texture features, BDIP and BVLC moments of the value component are chosen. For the selected features, we obtain multiresolution feature vectors which are extracted from all decomposition levels in wavelet domain. The multiresolution feature vectors of the color and texture features are efficiently combined by the normalization depending on their dimensions and standard deviation vector, respectively, vector components of the features are efficiently quantized in consideration of their storage space, and computational complexity in similarity computation is reduced by using progressive retrieval strategy. Experimental results show that the proposed method yields average $15\%$ better performance in precision vs. recall and average 0.2 in ANMRR than the methods using color histogram color autocorrelogram SCD, CSD, wavelet moments, EHD, BDIP and BVLC moments, and combination of color histogram and wavelet moments, respectively. Specially, the proposed method shows an excellent performance over the other methods in image DBs contained images of various resolutions.

• Proceedings of the IEEK Conference
• /
• 2001.09a
• /
• pp.669-672
• /
• 2001

We propose an efficient algorithm for an interactive content-based image retrieval using relevance feedback. In the proposed algorithm, a new query feature vector first is yielded from the average feature vector of the relevant images that is fed back from the result images of the previous retrieval. Each component weight of a feature vector is computed from an inverse of standard deviation for each component of the relevant images. The updated feature vector of the query and the component weights are used in the iterative retrieval process. In addition, the irrelevant images are excluded from object images in the next iteration to obtain additional performance improvement. In order to evaluate the retrieval performance of the proposed method, we experiment for three image databases, that is, Corel, Vistex, and Ultra databases. We have chosen wavelet moments, BDIP and BVLC, and MFS as features representing the visual content of an image. The experimental results show that the proposed method yields large precision improvement.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.27 no.3A
• /
• pp.258-267
• /
• 2002

We present an image retrieval method for improving retrieval performance by effective fusion of entropy features in wavelet region and wavelet moments. In this method, entropy features are sensitive to the local variation of gray level and well extract valley and edges. These features are effectively applied to contend-based image retrieval by well fusing to wavelet moments that represent texture property in multi-resolution. In order to evaluate the performance of the proposed method. We use Corel Draw Photo DB. Experiment results show that the proposed yields 11% better performance for Corel Draw Photo DB over wavelet moments method.

• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.1-17
• /
• 2017

The deep learning framework is software designed to help develop deep learning models. Some of its important functions include "automatic differentiation" and "utilization of GPU". The list of popular deep learning framework includes Caffe (BVLC) and Theano (University of Montreal). And recently, Microsoft's deep learning framework, Microsoft Cognitive Toolkit, was released as open-source license, following Google's Tensorflow a year earlier. The early deep learning frameworks have been developed mainly for research at universities. Beginning with the inception of Tensorflow, however, it seems that companies such as Microsoft and Facebook have started to join the competition of framework development. Given the trend, Google and other companies are expected to continue investing in the deep learning framework to bring forward the initiative in the artificial intelligence business. From this point of view, we think it is a good time to compare some of deep learning frameworks. So we compare three deep learning frameworks which can be used as a Python library. Those are Google's Tensorflow, Microsoft's CNTK, and Theano which is sort of a predecessor of the preceding two. The most common and important function of deep learning frameworks is the ability to perform automatic differentiation. Basically all the mathematical expressions of deep learning models can be represented as computational graphs, which consist of nodes and edges. Partial derivatives on each edge of a computational graph can then be obtained. With the partial derivatives, we can let software compute differentiation of any node with respect to any variable by utilizing chain rule of Calculus. First of all, the convenience of coding is in the order of CNTK, Tensorflow, and Theano. The criterion is simply based on the lengths of the codes and the learning curve and the ease of coding are not the main concern. According to the criteria, Theano was the most difficult to implement with, and CNTK and Tensorflow were somewhat easier. With Tensorflow, we need to define weight variables and biases explicitly. The reason that CNTK and Tensorflow are easier to implement with is that those frameworks provide us with more abstraction than Theano. We, however, need to mention that low-level coding is not always bad. It gives us flexibility of coding. With the low-level coding such as in Theano, we can implement and test any new deep learning models or any new search methods that we can think of. The assessment of the execution speed of each framework is that there is not meaningful difference. According to the experiment, execution speeds of Theano and Tensorflow are very similar, although the experiment was limited to a CNN model. In the case of CNTK, the experimental environment was not maintained as the same. The code written in CNTK has to be run in PC environment without GPU where codes execute as much as 50 times slower than with GPU. But we concluded that the difference of execution speed was within the range of variation caused by the different hardware setup. In this study, we compared three types of deep learning framework: Theano, Tensorflow, and CNTK. According to Wikipedia, there are 12 available deep learning frameworks. And 15 different attributes differentiate each framework. Some of the important attributes would include interface language (Python, C ++, Java, etc.) and the availability of libraries on various deep learning models such as CNN, RNN, DBN, and etc. And if a user implements a large scale deep learning model, it will also be important to support multiple GPU or multiple servers. Also, if you are learning the deep learning model, it would also be important if there are enough examples and references.