Ryuichi Oka

A method of model improvement for spotting recognition of gestures using an image sequence

We have developed a real-time gesture recognition system whose models can be taught by only one instruction. Therefore the system can adapt to new gesture performer quickly but it can not raise the recognition rates even if we teach gestures many times. That is because the system could not utilize all the teaching data. In order to cope with the problem, averages of teaching data are calculated. First, the best frame correspondence of the teaching data and the model is obtained by Continuous DP. Next the averages and variations are calculated for each frame of the model. We show the effectiveness of our method in the experiments.

A segmentation-free biometric writer verification method based on continuous dynamic programming

By introducing the continuous dynamic programming (CDP) algorithm developed by Oka [Oka, R., 1998. Spotting method for classification of real world data. The Comput. J. 41 (8), 559-565], we have developed a new segmentation-free, text-independent biometric writer verification method improving the average correct acceptance rate and the average correct rejection rate of forgeries to a practically useful level of about 95% and 100%, respectively, outperforming the only other biometric text-independent writer verification method of Yamazaki and Komatsu [Yamazaki, Y., Komatsu, N., 1999. Extraction of personal features from stroke shape, writing pressure and pen inclination in ordinary characters. In: Proc. 5th Internat. Conf. on Document Analysis and Recognition (ICDAR1999), pp. 426-429] based on a P-type Fourier descriptor-based learning vector quantisation scheme. To implement a text-independent, DP-based biometric writer verification, our spotting-enabling CDP algorithm has not only successfully traced and tracked an optimal microscopic dynamic features of real-time writing processes of either scanned characters or images of relevant individuals but also has implemented extension to a more secure text-independent writer verification method by exploiting the built-in spotting function of the CDP method, which is capable of ignoring inputs outside the task domain. This is so because a text-independent sample can be selected from any of many possible candidates located at an arbitrary location within the specified long reference sample. We have demonstrated the applicability of the method not only to Japanese text but also to English text that tends to assess the language independence of the method.

Non-monotonic Continuous Dynamic Programming for Spotting Recognition of Hesitated Gestures from Time-Varying Images

Continuous Dynamic Programming (CDP) has been proposed to recognize the meanings of human gestures from motion images. And this CDP has been extended to Non-monotonic CDP in order to recognize hesitated gestures. In this paper, we show the character of Non-monotonic CDP in detail.

Optimal Pixel Matching between Images

A two-dimensional continuous dynamic programming (2DCDP) method is proposed for two-dimensional spotting recognition of images. Spotting recognition is simultaneous segmentation and recognition of an image by optimal pixel matching between a reference and an input image. The proposed method performs optimal pixel-wise image matching and two-dimensional pixel alignment, which are not available in conventional algorithms. Experimental results show that 2DCDP precisely matches the pixels of non-linearly deformed images.

A mining method for linked Web pages using associated keyword space

We propose a novel method for mining knowledge from linked Web pages. Unlike most conventional methods for extracting knowledge from linked data, which are based on graph theory, the proposed method is based on our associated keyword space (ASKS), which is a nonlinear version of linear multidimensional scaling (MDS), such as quantification method type IV (Q-IV). We constructed a three-dimensional ASKS space using linked HTML data from the World Wide Web. Experimental results confirm that the performance of ASKS is superior to that of Q-IV for discriminating clusters in the space obtained. We also demonstrate a mining procedure realized by 1) finding subspaces obtained in terms of logical calculations between subspaces in an ASKS space and 2) detecting emerging spatial patterns with geometrical features

Spotting recognition and tracking of a deformable object in a time-varying image using two-dimensional continuous dynamic programming

When the background is varying or a tracking target is deforming, a unified processing composed of segmentations, recognitions and tracking is difficult to realize using conventional methods. An iterative method is proposed by a combination of spotting recognition with a reference image for tracking a target and making a segmented image as a reference image for the next frame. Simultaneous processing of recognition and segmentation is called spotting recognition. 2DCDP performs spotting recognition of images. An extended 2DCDP is used to deal with multiple object images with arbitrary shapes. Experimental results showed that the method worked well.

A segmentation-free method for image classification based on pixel-wise matching

Categorical classification for real-world images is a typical problem in the field of computer vision. This task is extremely easy for a human due to our visual cortex systems. However, developing a similarity recognition model for computer is still a difficult issue. Although numerous approaches have been proposed for solving the tough issue, little attention is given to the pixel-wise techniques for recognition and classification. In this paper, we present an innovative method for recognizing real-world images based on pixel matching between images. A method called two-dimensional continuous dynamic programming (2DCDP) is adopted to optimally capture the corresponding pixels within nonlinearly matched areas in an input image and a reference image representing an object without advance segmentation procedure. Direction pattern (a set of scalar patterns based on quantization of vector angles) is made by using a vector field constructed by the matching pixels between a reference image and an input image. Finally, the category of the test image is deemed to be that which has the strongest correlation with the orientation patterns of the input image and its reference image. Experimental results show that the proposed method achieves a competitive and robust performance on the Caltech 101 image dataset.

Image classification based on segmentation-free object recognition

This paper presents a new method for categorical classification. A method called two-dimensional continuous dynamic programming (2DCDP) is adopted to optimally capture the corresponding pixels within nonlinearly matched areas in an input image and a reference image representing an object without advance segmentation procedure. Then an image can be converted into a direction pattern which is made by matching pixels between a reference image and an input image. Finally, the category of the test image is deemed to be that which has the strongest correlation with the learning images. Experimental results show that the proposed method achieves a competitive performance on the Caltech 101 image dataset.

Full Pixel Matching between Images for Non-linear Registration of Objects

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.

Song wave retrieval based on frame-wise phoneme recognition

We propose a song wave retrieval method. Both song wave data and a query wave for song wave data are transformed into phoneme sequences by frame-wise labeling of each frame feature. By applying a search algorithm, called Continuous Dynamic Programming (CDP), to these phoneme sequences, we can detect a set of similar parts in a song database, each of which is similar to a query song wave. Song retrieval rates hit 78% in four clauses from whole databases. Differences in each query from song wave data and speech wave data is investigated.