Sakuichi Ohtsuka

Apparent 3-D image perceived from luminance-modulated two 2-D images displayed at different depths

An apparent 3-D image can be perceived from only two 2-D images displayed at different depths, when an observer views them from the direction in which they are overlapped. The two 2-D images are created from an original 2-D image by dividing its luminance according to independently obtained depth information. Subjective test results show that (1) an apparent 3-D image is perceived and (2) the perceived depth continuously varies according to the change in luminance ratio between the two 2-D images.

Human walking animation based on foot reaction force in the three-dimensional virtual world

This paper introduces a method that can generate continuous human walking motion automatically on an arbitrary path in a three-dimensional (3D) modelled scene. The method is based on a physical approach that solves the boundary value problem. In the motion generation stage, natural-looking walking motion, which includes plane walking, walking upstairs and downstairs and walking on a curved path, is created by applying dynamics and kinematics. The human body is approximated as a simple rigid skeleton model, and dynamic motion is created based on the ground reaction force of the human foot. To adapt to the 3D environment, the 3D walking path is divided into steps which are tagged with the parameters needed for motion generation, and step-by-step motion is connected end-to-end. Additional features include fast calculation and a reduced need for user control. The proposed method can produce interesting human motion and can create realistic computer animation scenes. Copyright

54.1: A Novel Direct-Vision 3-D Display Using Luminance-Modulated Two 2-D Images Displayed at Different Depths

We have found a novel three-dimensional (3-D) display mechanism using only two 2-D images displayed at different depths. It is based on a new perceptual phenomenon induced by the human binocular visual system and enables an observer using no extra equipment to perceive an apparent 3-D image of continous depth when the luminance is divided between the 2-D images according to the 3-D image depth. We have also develop a prototype direct-vision 3-D display using this mechanism. It can easily produce moving 3-D color images by using color conventional 2-D displays.

Dynamic shadow compensation of aerial images based on color and spatial analysis

This paper describes a method that can improve the visibility of features in shadowed regions in urban aerial images while retaining nonshadowed regions and the natural tint of shadowed regions automatically. The method first calculates the posteriori probabilities of each pixel belonging to shadowed class or nonshadowed class using color features from a low-frequency component image, and then compensates the intensity and saturation of each pixel in shadowed regions according to the posteriori probabilities. Experimental results indicate the effectiveness of proposed method.

53.3: A New Method for Protruding Apparent 3-D Images in the DFD (Depth-Fused 3-D) Display

We have found a new method in which two overlapped images displayed at different depths can be perceived as an image protruding outside the region of the two images. This enables a deeper 3-D imaging area in the DFD display, when the two images have opposite signs of luminance difference between an object in the images and the surroundings.

Linking images and keywords for semantics-based image retrieval

One of the major problems with existing content-based image retrieval systems is that their objective similarity can rarely match the users subjective and context-dependent similarity interpretation. We propose a novel approach to linking images and textual information to overcome this problem. First, salient image objects and also their structural and visual features are extracted. Next, keywords and images are linked in two stages: (1) by mapping low-level visual features of objects to related words using feature lexicons, and (2) by assigning words expressing higher-level semantic concepts to images on the basis of the feature-related words, lexical definitions, and the users relevance feedback. Experimental results show that the users retrieval semantics can be approximated better via this two-level multi-modality and also by supporting a large variety of querying and browsing schemes and thus higher-level interactivity.

Retrieval method for multi-category images

We propose an image retrieval technique based on multiple categories. Generally, an image can be associated with many concepts or objects. When concepts or objects are regarded as categories, an image could have multiple labels corresponding to these categories. Thus, image retrieval based on multiple categories would be an important technology for an image database system. To realize multicategory image retrieval, we employ the PMM (parametric mixture model) of multi-category text classification. We carried out multi-category image classification experiments using PMM and the color histogram feature. The experimental results show the possibility of multi-category image classification and retrieval using PMM.

Relating words and image segments on multiple layers for effective browsing and retrieval

This work proposes a new method for relating words and image segments by finding semantic coherence between these two cues on multiple layers. The method is based on the matching of visual segment clusters with words on various levels of abstraction. Our purpose here is to ease two main problems encountered in content-based image retrieval, namely, lack of semantic information captured by visual feature-based indexing and difficulty of handling subjectivity of user queries. The method is very promising for effective browsing and retrieval in large image data sets. It supports both target- and category-type browsing and searching schemes as well as textual and/or visual query specifications. Results of experiments on a wide, nonspecific image domain suggests that step by step semantic inference on consecutive layers of image-word association helps to improve accuracy of retrieval and browsing.

Human body posture measurement by matching foot pressure distribution to video images

We present a posture measurement method that uses foot pressure distribution and video camera images. The method first estimates the position, posture and size of subject in the real world from his or her foot pressure distribution. These estimates are represented by a stick model. The position of the skeleton in the camera image is then detected by using the estimated skeleton obtained by projecting the stick model onto the camera image as initial value. We tested this method on 80 walking patterns of eight women. The results show that the detected skeleton lies at the center of the leg.

Lower-body Posture Estimation of Walking People by Using Foot Pressure Distribution

This paper describes a method that uses video imaging and foot pressure distribution to estimate a persons lower body posture when walking. The method first estimates the position, posture, and size of the subject, from his or her foot pressure distribution.These estimates are represented by stick models, which are then projected onto video images to estimate the skeletal position. This method was tested on 90 walking patterns of 9 women, and the results show that most of the estimated skeletal structure lies inside the leg, thus confirming that this method functions effectively by repositioning the structure to the center of the leg and matching it to the edge of the video image.