Image Processing Laboratory

Ryuichi Oka Professor	Mohamed Hamada Associate Professor	Keitaro Naruse Assistant Professor	Yu Nakajima Assistant Instructor

First and foremost, the Image Processing Laboratory engages in research and development of image-based pattern recognition including some areas of Artificial Intelligence, database organisation, retrieval, and robotics. More specifically, as can be seen from the background information and the recent research publications of the members of the laboratory, 2-dimensional Continuous Dynamic Programming for non-linear image registration without mode, object tracking in a time-varying image, image retrieval from video data, character recognition of telop in a video image and extracion of the Internet community are our current focus. Related to the recent progress of the web, a huge amount of multimedia data without index becomes available to store in our PC's. However, no sophistcated methodology to manage such data has been developed so far. So that we are solicited to attach an index to each data. Our research aim is to develop algorithms to realize automatic annotation to real word data for integrated retrieval of multimedia information. The algorithms include self-organisation and transformation among representation of multimedia and feature extraction and recognition of real data. Real world data includes video, still image, speech, music, sound, and text each of which has not been indexed by labels. An example of multimedia integration retrieval called "CrossMediator" developed by ten-year project (1992-2002) RWC of METI (Japan). Some parts of CrossMediator have been in the commercial market by through a private company. Prof. R. Oka was a chief of group which developed "CrossMediator". Our laboratory will pursue to develop more sophisticated functions which might reveal a new generation of the Internet. Speech retreival system used in CrossMediator is used for developing new types of commercial products. At the present time we use two kinds of index for searching video data. One is used to obtain the target media from Youtube by using index. The other one is used for searching parts of the target data which are not indexed. One of the recent progress is developing a set of algorithms for spotting recognition and tracking objects in a time-varying image. These algorithms enable to realize automatic anonotation of video image capturing moving objects. The most promissing algorithm called 2-dimensional Continuous Dynamic programming (2DCDP) was proposed by our laboratory. This enables to perform full pixel matching between images. The full pixel correspondene is used for reconstructing 3D shape from at least 3 images obtained from different viewpoints and also gives a quite precise features like optical flow from a motion image. The 2DCDP is also applicable to non-linear registration for medical images to extract different parts between normal CT imagesand abnormal CT images. Another recent progress is developing a spotting algorithm for song wave retrieval. A large amount of song wave data is available for anyone along with the progress of storage hardware of music data. But the progress of technology for accessing the song wave data seems quite slow so that we must use conventional and not sophistcated tools. Our new software can provide users a convenient way to find and rerieve the song wave data by singing an arbitrary part of it which he/she wants to hear. The Internet is regarded as a network composed of virtual communities. Visualization of the community is becoming an important research target. Our lab is developing a sophistcated algorithm based on so-called Associated hyper-linked word space (A-space). We use the algorithm to detect clusters each of which corresponds a virtual community. Visualization of each virtual community is realized by showing the content extracted from the web pages included the community. On the other hand, we have carried out several research projects related to complex systems in which many independent components interact with each other and emerge global phenomena. Particularly, we have conducted on multiple robot control, full body motion generation in a humanoid, and analysis of human communication in the web space. With respect to multiple robot control, we have analyzed the dynamics of multiple mobile robots as a swarm under the mixture condition of long and short range communication, and evaluated the required amount of communication for the multiple robot control. With the analysis results, we have developed a control method of multiple mobile robots as a swarm. As implementing actual multiple mobile robots, we need a sensing system. In this project, we have developed a sensing and position estimating system for the robots. In terms of the full body motion generation in a humanoid, we have developed a method inspired from the mature gait, which is the change of walking pattern in human when growing from an infant through a child to an adult. For the analysis human communication in the web space, we have modeled a process of posting a message to the web space, and verified it with a large amount of actual web logs. Also, utilizing results drawing from the model, we have developed a new recommendation system considering both local interaction among users and emerged global phenomena.

Refereed Journal Papers

Refereed Proceeding Papers

[hamada-01:2009]	Mohamed Hamada. Pushdown Automata Simulator. Lecture Notes in Computer Science, 5670(2009):328–338, 2009.
	We study Pushdown Automata Simulator.
[hamada-02:2009]	Mohamed Hamada. Ants-like Agents: a Model and Analysis Based on Natural Ants Behavior. Lecture Notes in Artificial Intelligence, 5559(2009):20–29, 2009.
	We study a Model and Analysis of Based Ants-like Agents Based on Natural Ants Behavior
[hamada-03:2009]	H. El-Bakry and Mohamed Hamada. Fast Identification of Important Information in DNA Sequence by using High Time Delay Speed Neural Networks. Lecture Notes in Artificial Intelligence, 5712(2009):333–341, 2009.
	We study Fast Identification of Important Information in DNA Sequence by using High Time Delay Speed Neural Networks
[oka-01:2009]	Yuichi Yaguchi, Kenta Iseki, Nguyen Tien Viet, and Ryuichi Oka. Full Pixel Matching between Images for Non-linear Registration of Objects. IPSJ Transactions on Computer Vision and Applications, 2(1):1–14, 2010.
	A two-dimensional continuous dynamic programming (2DCDP) method is proposed for two-dimensional (2D) spotting recognition of images. Spotting recognition is the simultaneous segmentation and recognition of an image by optimal pixel matching between a reference image and an input image. The proposed method performs optimal pixel-wise image matching and 2D pixel alignment, which are not available in conventional algorithms. Experimental results show that 2DCDP precisely matches the pixels of nonlinearly deformed images.

[naruse-01:2009]	Keitaro Naruse. Stability Analysis of Flocking and Motion Control in Swarm Robot. In Proc. of the 9th Asia-Pacific Complex Systems Conference, 2009.
	When navigating a massively large number of robots as a group, which are called swarm robots in this paper, it is important to reduce an amount of communication while they are working. The problem discussed here is to navigate the swarm robots at a goal position when only a single robot (called a leader) is given the goal position and the other robots know it through communication. Then, the objective is to design a communication network among the swarm robots with a small amount of communication, while maintaining the response and speed of the robots for getting at the goal. For this problem and objective, each of the robots( called a follower ) is assumed to communicate each other in its neighbor, and some robots( called sub-leaders ) are allowed to communicate to the leader in a long distance beyond the neighbor. In this paper, the relation between the number of the sub-leaders and　followers and the response and speed to the goal is discussed by both an empirical method of computational simulations and an analytical method of a dynamical system of multiple particles. Also, the stability and condition of the flocking, which is defined as all of the robots are connected in a communication neighbor are analyzed. The main contribution of this paper is that that a control method for the swarm robots is given as a communication network of the robots, and the stable conditions and control parameters for the control method are also given.

[naruse-02:2009]	Keigo Suenaga and Keitaro Naruse. Design of Mobile Robot Network for Efficient Flocking. In Proc. of Dynamics and Design Conference 2009, 2009.
	When controlling a massively large number of mobile agents, the scalability to the number of the agents is one of the important issues. In this paper, we consider it as the motion control of flock in which a small number of leaders which know a goal navigate a large number of followers which do not. The follower determines its motion only interacting with geometrical neighbors. When we need more agents, the framework allows us to simply add followers with no adjustment or tuning, if the leaders navigate all the followers. In this paper, we analyze and discuss the flock control capacity by the leaders.

[naruse-03:2009]	Keitaro Naruse. Biped Walking Pattern by Virtual Muscle Oscillation in Growing Physical Parameter of Robot Model. In Proc. of ICROS-SICE International Joint Conference 2009, pages 2696 - 2699, 2009.
	In this paper, an approach for the biped walking pattern generation, inspired from a learning process of human walking from a child to an adult, is taken. When an infant grows up to a child, it gradually learns how to walk through trials and errors. Then, as a child grows up to an adult, the body size and weight become larger, and a walking pattern is changed from one of a child. The process is called mature gait. The objective of this paper is to analyze the nature of the mature gait and to investigate the applicability of the mature gait to generation of robot biped walking pattern. Out simulation results show that physical growth of the robot model helps speeding up a learning process of walking pattern generation.
[oka-02:2009]	Yuichi Yaguchi, Kenta Iseki, Nguyen Tien Viet, and Ryuichi Oka. 3D Object Reconstruction Using Full Pixel Matching. In Computer Analysis of Image and Pattern, pages pp.873–883. LNCS, September 2009.
	This paper proposes an approach to reconstruct 3D object from a sequence of 2D images using 2D Continuous Dynamic Programming algorithm (2DCDP) as full pixel matching technique. To avoid using both calibrated images and fundamental matrix in reconstructing 3D objects, the study uses the same approach with Factorization but aims to demonstrate the effectiveness in pixel matching of 2DCDP compared with other conventional methods such as Scale-Invariant Feature Transform (SIFT) or Kanade-Lucas-Tomasi tracker (KLT). The experiments in this study use relatively few uncalibrated images but still obtain accurate 3D objects, suggesting that our method is promising and superior to conventional methods.
[oka-03:2009]	Jun Ma, Kenta Iseki, Yuichi Yaguchi, and Ryuichi Oka. Segmentation-Free Object Recognition Using Full Pixel Matching. In IEEE Nine International Conference on Computer and Information Technology, pages 283–288. IEEE, October. 2009..
	We present a novel method for recognizing anobject in an image using full pixel matching between a referenceimage and an input image without advance segmentation of theimage. A method called two-dimensional continuous dynamic programming( 2DCDP) is adopted to optimally calculate theaccumulated local distances of all corresponding pixels innonlinearly matched areas in an input image and a referenceimage representing an object. The object is recognized byusing two parameters, the matching rate and the standarddeviation of the amplitude of a vector of pixel displacementbetween matched pixels, so that images can be mapped intoa two-dimensional space. Finally a general decision space isproposed for nonlinear transformation in object recognition.Experimental results show that the proposed method perform swell in recognizing objects.
[oka-04:2009]	Takashi Wagatsuma, Yuichi Yaguchi, and Ryuichi Oka. Cross-Media Data Mining using Associated Keyword Space. In IEEE Nine International Conference on Computer and Information Technology, pages 289–294. IEEE, October. 2009.br>
	This paper proposes a method to analyze and determine the unified similarity of various data, such as movies, images, sound, and text. Items from these sources of data have various relative-similarity properties, which are specified by quantitative methods, and these similarities can be represented by physical distances in a multidimensional scaling space. In this study, we introduce an example of a 3-D multimedia space using the Associated Keyword Space (ASKS) and demonstrate similarity relationships between various sources of data in this space.
[oka-05:2009]	Hisato Aota, Yuichi Yaguchi, Kenta Iseki, and Ryuichi Oka. Extracting Motion Features of Objects Based on Full Pixel Matching. In IEEE Nine International Conference on Computer and Information Technology, pages 300–305. IEEE, October. 2009.
	This paper proposes an approach to extract motion features from sequences of images of human behavior. A novel algorithm called two-dimensional continuous dynamic programming (2DCDP) is proposed, which can obtain a set of correspondence data for all pixels between sequential images. The 2DCDP algorithm performs segmentationfree detection of objects in an input image from representations in reference image. The output of the 2DCDP algorithm describes a complete vector field that indicates detailed motion of objects. Experiments were performed to show the precision of he motion features extraction.
[oka-06:2009]	Yuichi Yaguchi, Yusuke Sakai, Keisuke Yoshida, and Ryuichi Oka. Web Video Data Clustering and Recognition using Histograms of Phoneme Symbols. In IEEE Nine International Conference on Computer and Information Technology, pages 306–311. IEEE, October. 2009.
	The clustering and recognition of Web video content play an important role in multimedia information retrieval. This paper proposes a method for both clustering and recognizing Web video content using a histogram of phoneme symbols (HoPS). HoPS contains information about speech and sound intervals. In this study, three experiments were conducted. The first experiment allocated HoPS feature of video intervals in a 3D space using PCA and quantification method IV (QIV). The second experiment applied the k-nearest neighbor (k- NN) method to analyze the difficulties in clustering. The third experiment recognized unknown video intervals by using the distance between HoPS of the query and a category average. The accuracy of the recognition results were 44.3% and 36.9% using the Mahalanobis distance and the correlation distance for the category average of training data, respectively. Keywords: Video content clustering, Phoneme recognition, Multimedia data mining.
[oka-07:2009]	Kazuhiro Ota, Yuichi Yaguch, and Ryuichi Oka. Recognition and tracking of moving objects using mean-shift and pixel. In Proceedings of meeting on Image Recognition and Understanding, pages 1356–1363. IEICE, July. 2009.
	The paper proposes a method composed of two stages for realizing both automatic identi cation of objects to be tracked in a time-varying image. The rst stage of the method uses so-called mean-shift algorithm for obtaining sub-regions each of which becomes candidate of object. The second stage of the method uses extracted sub-region as a reference image of so-called 2-dimensional Continuous Dynamic Programming for matching it to other frames in a time-varying image. Let sub-regions that are moving between frames be the tracking target, then tracking will be done by iteration of matching of the tracking target.
[oka-08:2009]	Yuichi Yaguchi, Hisato Aota, and Ryuichi Oka. Pixel Tracker. In Proceedings of meeting on Image Recognition and Understanding, pages 1237–1244. IEICE, July. 2009.
	This paper proposes a method called ”Pixel Tracking. ” Pixel tracking means succession of pixel correspondence between images. One time pixel correspondence is realized by so-called 2-D Continuous Dynamic Programming (2DCDP) by means of that each pixel of an image is optimally corresponding to a pixel of another image. All pixels of one time pixel correspondence make a vector field indicating a fine flow pattern by which we can extract many new kinds of feature like shape of object and change of shape etc from a time-varying image. These features is not easily obtained from those of conventional optical flow or SIFT tracker etc. Experimental results shows the comparison of tracking between the method of pixel tracking and the conventional methods by applying to different images. The experiments confirm the extraction of new features by means of pixel tracker.

Unrefereed Papers

[naruse-04:2009]	Keitaro Naruse. Navigation of Swarm Robots and Analysis of Flocking Stability. In Proc. of the 2009 SICE System and Infomation DivisionConference, 2009.
	The problem discussed in this paper is to navigate a large number of robots to a single destination point. Considering the scalability of the robots, it is assumed that only a single robot is given the destination, and all the other robots are required to communicate with neighboring robots for knowing the destination. Under the assumption, the robots should be always remained flocked for the communication and navigation. In this paper, it is defined as the flocking stability, it is discussed analytically, and some criteria for the stable flocking are shown.
[naruse-05:2009]	Luo Jie and Keitaro Naruse. Localization of Multiple Mobile Robots by Nonparametric Filter and Evaluation of Position Errors. In Proc. of the 19th Intelligent Systems Symposium, 2009.
	The objective of this paper is to develop a localization system for multiple mobile robots, in which each robot is assumed to observe a set of known landmarks. If only limited numbers of them are available, it suffers from law accuracy. For improving it, in this paper, it is assumed that a robot can know the other robots and their positions estimated by themselves and utilize the positions for its localization. In other words, the other robots can be viewed as new landmarks. However, since the other robots as the new landmarks can include some amount of position errors, the estimated position of the robot cannot be high enough. This paper investigates the localization errors of a robot when applying a nonparametric filter into multiple robot localization system.
[naruse-06:2009]	Keigo Suenaga and Keitaro Naruse. Study of Swarm Robots Control using Hierarchy Agent Model. In Proc. of the 19th Intelligent Systems Symposium, 2009.
	This paper present the development of the control model of the swarm robots called the swarm leading strategy. The model invloves the hierarchical agent model with respect to the communication function.
[naruse-07:2009]	Keitaro Naruse. Body growth and mature gait in musculoskeletal system. In Proc. of the 2009 Fall JSPE Conference, 2009.
	In this paper, an approach for the biped walking pattern generation, inspired from a learning process of human walking from a child to an adult, is taken. When an infant grows up to a child, it gradually learns how to walk through trials and errors. Then, as a child grows up to an adult, the body size and weight become larger, and a walking pattern is changed from one of a child. The process is called mature gait. The objective of this paper is to analyze the nature of the mature gait and to investigate the applicability of the mature gait to generation of robot biped walking pattern. Out simulation results show that physical growth of the robot model helps speeding up a learning process of walking pattern generation.
[naruse-08:2009]	Keitaro Naruse and Takuma Daicho. Landmark Recognition in Omnidirectional Vision by High Order Local Autocorrelation Features. In Proc. of the 2010 Spring JSPE Conference, 2010.
	In the localization problem of mobile robots, one of the important problems is the recognition of landmarks, which should be fast and accurate enough for the real time operation of the robots. In this paper, we consider the omnidirectional image taken from the mobile robot for the landmark detection and high order local autocorrelation (HLAC) features for the image processing of finding the distance. However, we have not understood fully how well HLAC features work for the landmark recognition of the omnidirectional images. In this paper, we discuss it in some practical situations.
[naruse-09:2009]	Keigo Suenaga and Keitaro Naruse. Change of Consensus Speed of Flock using Leader-Follower StrategyFeatures. In Proc. of the 2009 JSME Robotics and Mechatronics Conference, 2009.
	In flock control using a Leader-Follower strategy, we discuss about a change of the consensus speed of the flock by the numerical change of followers observing a leader. With one leader, the number of followers that leader can control is limited. However, it can be settled by long-range communication follower agent, Sub-Leader. In the proposal method (200 agents flock case), the number of the necessary Sub-Leader agents was 25%.
[naruse-10:2009]	Masao Kubo, Keitaro Naruse, Hiroshi Satoh, and Takashi Matsubara. Population Estimation from Posted Article Distribution by Queuing theory. In Proc. of the 19th Intelligent Systems Symposium, 2009.
	This paper presents the modeling of the social interaction of users in the Web forums, and the estimation method of the population size of the Web forums using the developed model.
[naruse-11:2009]	Keitaro Naruse, Hiroki Matsuura, Shinya Mizoe, Kazuhiro Ota, Yusuke Sakai, Keigo Suenaga, Takashi Wagatsuma, and Luo Jie. Reports on Studio Course of Mobile Robots Development in Univ. of Aizu. In Proc. of the 2009 SICE System Integration Division Conference, 2009.
	This report presents the studio course of developing mobile robots in the graduate school of university of Aizu. Each robot equips with an omni-directional camera, and it estimates its location using a set of landmark in the workspace, as well as communicating with the other robots to know where they are. The experiments show the robots avoid collision to the others by enough accuracy of localization.
[oka-09:2009]	Yuichi Yaguchi, Marie Oshima, Takashi Kawai, and Ryuichi Oka. Kansei Information Retrieval from Web Video Repository. In Technical Report of PRMU2009-11, pages 59–64. IEICE, May 2009.
	This paper proposes and reports pilot study of Kansei information retrieval using Web based speech retrieval system for Youtube video repository. Normally, video data uploaded on the Web has many characteristics and specifications. Thus, retrieval system on the Web need generalized method rather than conventional one. In this paper, we try to retrieve similar video segment of query word for create Kansei histogram of query that is using AiZoom which is a speech retrieval system using frame-wise phoneme recognition forWeb video data. Additionally, we inspect optical flow of these similar video segment for getting Kansei feature.
[oka-17:2009]	Keigo Suenaga, Keitaro Naruse, and Masao Kubo. Control of multi-agent swarm by Boids model. In Proceedings of JSME Robotics and Mechatronics Conference 2008, 2008.

Academic Activities

[hamada-04:2009]

H. El-Bakry and Mohamed Hamada. A New Implementation for Neural Networks in Fourier-Space, pages 70–92. Number 1-2009 in Foundations in Computational Intelligence. Springer-Verlag, Germany, 2009.

Academic Activities

[hamada-05:2009]	Mohamed Hamada, April 2009. IEEE Member.
[hamada-06:2009]	Mohamed Hamada, April 2009. ACM Member.
[naruse-12:2009]	Keitaro Naruse, 2009. Reviewer, IEEE International Conference on Computer and Information Technology.
[naruse-13:2009]	Keitaro Naruse, 2009. Session Organizer and Session Chair, SOFT Intelligent System Symposium 2009.
[naruse-14:2009]	Keitaro Naruse, 2009. Session Organizer and Session Chair, SICE System Integration Division Conference.
[naruse-15:2009]	Keitaro Naruse, 2009. General Co-Chair and Local Arangement Chair, SOFT Intelligent System Symposium 2009.
[oka-18:2009]	Ryuichi Oka, May 2009. Reviewer of submitted papers, The Acoustic Society of Japan.
[oka-19:2009]	Ryuichi Oka, May 2009. Reviewer of submitted papers, The Institute of Electronics, Information and Communication Engineers.
[oka-20:2009]	Ryuichi Oka, April 2009. Senior Member of editorial committee, Japanese Society for Artificial Intelligence.
[oka-21:2009]	Ryuichi Oka, April 2009. Reviwer of submitted papers, The nine Inter. Conf. on Computer and Information Technology.

Ph.D., Master and Graduation Theses

[hamada-07:2009]	Sayota Sato. Graduation Thesis: Simulator and Robot Game for Learning Automata Theory, University of Aizu, 2009. Thesis Advisor: M. Hamada
[hamada-08:2009]	Minoru Takeda. Graduation Thesis: Learning Style Index: Extension and Implementation, University of Aizu, 2009. Thesis Advisor: M. Hamada
[naruse-16:2009]	Jun Kinjo. Graduation Thesis: The color conversion of the color image which can be recognized by a color blindness, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-17:2009]	Kazuya Shoji. Graduation Thesis: The Sequentiality of Utterance Time in Interaction, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-18:2009]	Reiko Toma. Graduation Thesis: Text message analysis by foucsing on emotion, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-19:2009]	Kyoko Tsugioka. Graduation Thesis: Silent Speech Recognition by Using Higher-order Local Auto-Correlation, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-20:2009]	Yukihiro Daido. Graduation Thesis: The Iterated Prisoner’s Dilemma game and Transforming Society, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-21:2009]	Takuya Senjo. Graduation Thesis: Dynamic Stability of Passive Dynamic Walker with Resilient Joint, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-22:2009]	Shigekazu Fukui. Graduation Thesis: Semantics Extraction from Text by Statistical Means, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse
[naruse-23:2009]	Takuma Daicho. Graduation Thesis: Object detection from an omnidirectional image using higher order local auto correlation feature, University of Aizu, 2010. Thesis Advisor: Keitaro Naruse

Others

[oka-22:2009]

Ryuichi Oka, Yuichi Yaguchi, and Kenta Iseki. Application of Japanese patent titled Image pattern matching device, and method and program, Tokugan-2009-5682, 2009, 14-th, January, 2009

2011 THE UNIVERSITY OF AIZU ALL RIGHTS RESERVED