Multispectral images, which consist of more color components than RGB images, are expected to be used for vegetation analysis and medical imaging. A capturing system with multispectral filter array technology has been researched to shorten the taking time and reduce the cost. In this system, the mosaicked image captured by the multispectral filter array is demosaicked to reconstruct the multispectral image. In this paper, we propose the demosaicking method using decorrelated vectorial total variation (D-VTV) regularization for a multispectral image. First, the demosaicking process is regarded as inverse problem of the image observation model. Then, the reconstructed image is estimated by minimizing D-VTV as a regularization term under the constraint condition. In the experimental results, the quality of the reconstructed image by the proposed method is better than that of the reconstructed image by a conventional approach.
In quantitatively evaluating the contour shapes, there are a number of methods using Fourier descriptors. In particular there is a method with elliptic Fourier descriptors (EFds) as one method suitable for the analysis of the contour shapes. EFds are used for analytically parsing closed, two-dimensional contours and have been applied to evaluation of biological shapes. In this paper, we quantitatively evaluate contour shapes of clothing by using EFds and we propose a classification method of clothing shapes. We apply the EFds in binarized image of clothing and reduce the number of dimensions of multivariate Fourier coefficients by performing a Principal Component Analysis (PCA). We classify principal component scores obtained from features by Support Vector Machine (SVM). We classified clothing into three types and compared principal component score of EFds with BoF vector of SIFT. From the results, we show the effectiveness of EFds in the classification of clothing contour shapes. By using multivariate Fourier coefficients obtained from EFds, it is possible to constitute and express visual contour shapes of clothing.
In brainstorming, participants have to keep some important rules; (i) do not deny any proposals, (ii) do propose more proposals. Therefore, the evaluation by the participant for proposals can be represented by the non-verbal behaviors by participants. In this report, we propose image processing methods for the detection of non-verbal behaviors by participants. Moreover, we propose an evaluation method for the activity of the brainstorming. Finally, we discuss the relationship between the proposed activity and the review result by the participant.
Not only verbal information but also non-verbal information plays an important role for impression formation at human-to-human communication. In this work, taking the facial region movement as non-verbal information, we study the relationship among them. DSCQS (Double Stimulus Continuous Quality Scale) method is used as the subjective experiments. Nose regions are easily processed and horizontal and vertical movements of the nose are detected. Finally, a positive co-relation is detected between DSCQS evaluation and the facial vertical movement.
Both KQML and FIPA ACL are the agent communication language which has been designedbasedontheSpeechActTheory torealizeintelligentPoint−to−Pointcommunicationprotocol. AnideatoapplytheframeworkofFIPAACLtoSpeechCodeTheorywhichhasbeenderivedfromSpeech ActTheoryhasbeenthought andthepossibilitytohavetherelationshiptointerculturalcommunication has been considered.
Scalable Poisson disk sampling technique is one of the most powerful NPR techniques to generate artistic images from input photos. It can generate various kinds of NPR images such as colored paper mosaic, pointillistic painting and so on. However, there has been a limitation that users cannot edit parts of generated NPR images. In this paper, we propose an interactive editing method of scalable Poisson disk sampling distribution. By applying our method, users can interactively modify the artistic image and generate the variegated NPR images.
When we convert a color image to a monochrome image, we have to reduce information quantity as three base colors to single base color. Usually this reduction is achieved with luminance information. Since human receive more information from luminance channel compared to color differences, luminance image is perceptually good. However in a colorful image, many colors converted into the same luminance and we could not distinguish them. In this research, we use color difference information in monochrome conversion to handle such colorful images.
This paper presents 3 types of motion capture systems from video cameras. These systems can capture ① movements of intertwisting wrestlers，② walking of pedestrians outdoor and ③ movements of bodies such as arms and legs as well as the finger movement and facial expression simultaneously without any marker and sensor. The video camera also records speech of person. HMM (Hidden Markov Model) can produce speaking gesture according to the scenario based on symbolic relationship between the speech and action. Moreover, we consider the most suitable action to the contents of speech.