Face detection is a hot research topic in Computer Vision; the field has greatly progressed over the past decade. However, to our knowledge, face detection in low-resolution images has not been studied. In this paper, we use a conventional AdaBoost-based face detector to show that the face detection rate falls to 39% from 88% as face resolution decreases from 24×24 pixels to 6×6 pixels. In order to improve the face detection rate when using low-resolution images, we propose a new face detection method comprising four techniques. 1. Using an upper-body detector. 2. Expansion of input images. 3. Frequency-band limitation of features. 4. Combination of two detectors. As a result, our method improved the face detection rate from 39% to 71% for 6×6 pixel faces of MIT+CMU frontal face test set.

View full abstract

The development of advanced image compression techniques is demanded as the chance to communicate high detailed digital images, and to do the storage maintenance increases rapidly. This paper presents a novel scattered point interpolation technique called MCA (Multiresolution CSRBF Approximation) that is composed of multiresolution analysis of wavelet transform and CSRBF. The lossy image coding system for compressing function information of CSRBF approximation surface based on MCA is suggested. We showed the superiority of MCA by evaluating the accuracy of scattered point detection, and the accuracy of scattered point interpolation. Then, the compression performance of the proposed coding system was compared with JPEG. Our results demonstrated that the proposed approach has better quality than JPEG in low bit rate.

View full abstract

Active Contour Model, Snakes, is one of the most popular boundary detection methods. Snakes has a problem that it is difficult to determine an initial contour and parameters suitable for the target shape. In addition, this method cannot detect the target boundary precisely when the target image does not have clear edges or uniform feature in the target domain and the background domain. This paper proposes Snakes with decentralized cooperative processing. In this method, multiple Snakes do the detection procedure in a parallel and cooperative manner. As a result, the possibility of obtaining the target contour is increased by exchanging information between the multiple Snakes. This paper also presents effectiveness of this approach through the experiments with simulation images and real images.

View full abstract

This paper proposes a template matching method to search a target image area that has arbitrary location and inclination. The geometrical parameters that represent a contour form in a local image area are used for matching keys. An inclined target image area is searched in the voting space that is accumulated by votes based on the geometrical parameters. The proposed method realizes high-speed processing and less memory resources by integrating local image features. The experimental results show that the target image areas that had free locations and free inclinations were searched successfully with less computational cost by the proposed method.

View full abstract

When a display area is limited like mobility or ubiquitous environment, there is a problem of hiding windows when there are a lot of windows. The focus switch is necessary to refer to the hidden window. This can be said that the context of work is discontinuous. In this paper, we propose a window operation interface with context continuity. First of all, to keep the continuous context, use of user's head movements is examined. Next, typical head movement of usually work is measured with 3-D magnetic positional sensor. A method of distinguishing intentional movement and other movement is examined. A window positional control method is designed based on the results of examination. And, an evaluation experiment of this control method is conducted and effectiveness is shown. Finally, a method of face recognition by camera which don't use tethered sensor is examined, and evaluated.

View full abstract

This paper proposes a semi-automatic segmentation method for the 3D-ISM system which enables the capture of a high-resolution full-color cross-sectional image series of biological sample. Our approach is based on region-based segmentation and an adaptive classification technique by using the Otsu method, so it can be applied to an object like biological tissue which has different features by location. As a result, we have achieved to develop the method to decrease the degree of manual operation required. This paper also shows experimental results of applying our method to visualize the internal structure of the equine ovary. We have confirmed the spatial arrangement inside the ovary which had not been revealed so far.

View full abstract

Edge detection is the process of characterizing object boundaries and therefore a problem of fundamental importance in image analysis. In general, the development of edge detection commonly uses a multiple stage algorithm such as reducing noise, finding the intensity gradient and tracing edges through the image to detect wide range of edges. While this set of algorithms can produce a significant advantage over the conventional edge detections, the problem of selecting an appropriate parameter on each stage has been taken into account since most of parameters have to be determined interactively referring to the input images. In this study, we proposed a novel high performance edge detection based on regression surface with a single parameter for luminance images. The proposed method fits regression surfaces to the luminance images to reduce noise and calculate the intensity gradient. The parameter to control the resulted edges is a threshold of the non-edge's percentage. This parameter can be set simply with edge and non-edge's histogram as reference. Our evaluation shows that the proposed edge detection with a parameter fixed to basic and photographic images has significant advantage over the Canny operator in which parameters are adaptively adjusted to those.

View full abstract