Toshinori Watanabe

A new pattern representation scheme using data compression

We propose the PRDC (Pattern Representation based on Data Compression) scheme for media data analysis. PRDC is composed of two parts: an encoder that translates input data into text and a set of text compressors to generate a compression-ratio vector (CV). The CV is used as a feature of the input data. By preparing a set of media-specific encoders, PRDC becomes widely applicable. Analysis tasks - both categorization (class formation) and recognition (classification) - can be realized using CVs. After a mathematical discussion on the realizability of PRDC, the wide applicability of this scheme is demonstrated through the automatic categorization and/or recognition of music, voices, genomes, handwritten sketches and color images.

Fast agglomerative hierarchical clustering algorithm using Locality-Sensitive Hashing

The single linkage method is a fundamental agglomerative hierarchical clustering algorithm. This algorithm regards each point as a single cluster initially. In the agglomeration step, it connects a pair of clusters such that the distance between the nearest members is the shortest. This step is repeated until only one cluster remains. The single linkage method can efficiently detect clusters in arbitrary shapes. However, a drawback of this method is a large time complexity of O(n2), where n represents the number of data points. This time complexity makes this method infeasible for large data. This paper proposes a fast approximation algorithm for the single linkage method. Our algorithm reduces the time complexity to O(nB) by rapidly finding the near clusters to be connected by Locality-Sensitive Hashing, a fast algorithm for the approximate nearest neighbor search. Here, B represents the maximum number of points going into a single hash entry and it practically diminishes to a small constant as compared to n for sufficiently large hash tables. Experimentally, we show that (1) the proposed algorithm obtains clustering results similar to those obtained by the single linkage method and (2) it runs faster for large data than the single linkage method.

Swarming robots-foraging behavior of simple multirobot system

Research of multirobot systems is active in these days. The most remarkable characteristic of multirobot system is that the robots work cooperatively and achieve the task which a single robot cannot do. This characteristic is important and it is essential to investigate an effect of number of robots in the system. There are some tasks which art! suitable for multirobot system and we chose foraging task. A robot used in this paper has a simple interaction method with a light signal. In this paper, we discuss their behavior in a field that has one feeding point at first. Their behavior is analyzed by a robot simulation and a mathematical model. Next, we discuss a performance of the system in a clockface arranged foraging field, in which some localized foods are located equidistant from the home. Here, we report the group shows an ordered behavior depending on the number of robots and the strength of the interaction.

Motion Vector Based Moving Object Detection and Tracking in the MPEG Compressed Domain

As MPEG standards prevail, the opportunities to handle MPEG compressed videos increase, and the video indexing and management that can directly process the compressed videos become important. MPEG video coding standards use motion compensation to compress video data, and the motion compensation generates motion vectors that contain motion information similar to optical flows between regions in different frames. Although motion vectors are useful for video analysis, they are not always generated along moving objects, and it is difficult to analyze moving objects using only these vectors. In this paper, we propose a moving object detection and tracking method in the MPEG compressed domain for video surveillance and management. In our method, we introduce images that record moving regions and accumulate unmoving regions in which the moving objects are expected to exist after the current frame. By utilizing these images, we can detect and track moving objects using only motion vectors even if the motion vectors of moving objects become zero vectors due to their behaviors and are lost due to their picture type. We demonstrate the effectiveness of the proposed method through several experiments using actual videos acquired by an MPEG video camera.

Traffic-like Movement on a Trail of Interacting Robots with Virtual Pheromone

Study on traffic flow is an attractive topic from physical viewpoint as well as from engineering viewpoint, and many researchers have studied it. Some of them have proposed a cellular-automata model of traffic flow on a trail, and showed some interesting results. Chowdhury et al. have proposed a cellular automata model motivated by the motion of ants with using pheromone, and reported that there is a critical density of elements where the traffic flow increases drastically under a certain condition. In this paper, we discuss the robustness of this phenomenon using real robot system with light sensors and device to simulate chemical signals by computer graphics.

Object Discovery by Clustering Correlated Visual Word Sets

This paper presents a novel approach to discovering particular objects from a set of unannotated images. We aim to find discriminative feature sets that can effectively represent particular object classes (as opposed to object categories). We achieve this by mining correlated visual word sets from the bag-of-features model. Specifically, we consider that a visual word set belongs to the same object class if all its visual words consistently occur together in the same image. To efficiently find such sets we apply Min-LSH to the occurrence vector of the each visual word. An agglomerative hierarchical clustering is further performed to eliminate redundancy and obtain more representative sets. We also propose a simple and efficient strategy for quantizing the feature descriptors based on locality-sensitive hashing. By experiment, we show that our approach can efficiently discover objects against cluster and slight viewpoint variations.

A New Stable AQM Algorithm Exploiting RTT Estimation

AQM is a technique for congestion control such that a router notifies congestion to a TCP sender when congestion occurs. Almost no AQM algorithms ever take the RTT values of TCP connections into account in congestion control, despite they are essential parameters. This paper proposes a new AQM algorithm that exploits them explicitly by introducing a passive RTT estimation technique in a router. The simulation results show that our AQM stabilizes the queue length better than previous AQM algorithms

Similarity-based image retrieval system using partitioned iterated function system codes

We propose a new image retrieval system using partitioned iterated function system (PIFS) codes. In PIFS encoding, a compression code contains mapping information between similar regions in the same image. This mapping information can be treated as vectors, and representative vectors can be generated using them. Representative vectors describe the features of the image. Hence, the similarity between images is directly calculable from representative vectors. This similarity is applicable to image retrieval. In this article, we explain this scheme and demonstrate its efficiency experimentally.

Fast Hierarchical Clustering Algorithm Using Locality-Sensitive Hashing

A hierarchical clustering is a clustering method in which each point is regarded as a single cluster initially and then the clustering algorithm repeats connecting the nearest two clusters until only one cluster remains. Because the result is presented as a dendrogram, one can easily figure out the distance and the inclusion relation between clusters.

Robust automatic video object segmentation with graphcut assisted by SURF features

Video object segmentation is a task to distinguish the foreground from the background in videos. Most previous research on automatic video object segmentation based on graphcut segmentation uses the motion cue and the color cue to separate the background from the foreground. Consequently, the segmentation result deteriorates when the motion and/or the color becomes disordered, which typically occurs when a moving object stops and when a light is switched on/off. This paper proposes a new automatic video segmentation method robust to unstable motion and color. To achieve robustness, the graphcut segmentation is supported by the SURF feature, which is highly invariant to the change of scale, rotation, and luminance. In particular, our method matches the SURF features between two consecutive frames and modifies the segmentation result when the matched SURF features are assigned different labels.