In order to achieve color management for open systems such as Internet, it is important to elucidate the color characteristics of individual equipment. Traditionally, printed test charts have been employed to evaluate the color reproduction of digital cameras. But this evaluation method could not measure the spectral responsivity which is the most fundamental color characteristics of digital cameras. Color signal outputs from digital cameras can be calculated from spectral distribution of an illumination, spectral reflectance of a shooting object, and spectral responsivity of a camera. The methods of measuring spectral distribution of illuminations and spectral reflectance of objects have been established unambiguously, and their characteristics are available from various databases. However, no accurate methods have been clearly defined regarding the measurement of the spectral responsivity characteristics of digital cameras. This paper describes the measurement method which realize to measure the spectral responsivity including negative lobes. As the subject of our new measurement method, we assume a digital camera which simply adopt linear matrix for color signal processing. The newly developed methods can reduce the effects of measurement errors caused by automatic functions such as automatic exposure controlling or automatic white balancing which are adopted by the majority of consumer digital cameras. This paper also describes an arrangement of equipment, definition of test chart and raw data handling together with some worked examples. The International Standard based on our methods has been published as IEC 61966-9.

View full abstract

We propose a new recognition method that combines the distance information shown by the slipping between an input pattern and reference pattern and the structural information shown by the dissimilarity related to the relative stroke position obtained from the input pattern and a standard. The discrimination is performed by using the weighed mean of the two kinds of information. This method is based on the evidence that the two kinds of information have high independence, which is clarified by the newly derived matrix form to describe the distance and the structure commonly. In the recognition experiment for free-style handwritten characters, the proposed method achievers higher performance than the method using just the distance information, and its effectiveness is confirmed. Moreover, the result shows the structure information plays a complementary role of recovering the correct category when the distance information itself cannot rank the correct category in high order because the character degradation is large.

View full abstract

This paper proposes a new method to recover the sign of local surface curvature of object with multiple reflectance from multiple shading images using neural network. This method uses a calibration sphere to classify the surface by curvature. The previous approaches using a calibration sphere assume that the reflectance of the object is constant. Even if the object has varying reflectance, this method can get the robust results by using multiple shading images (more than three) and Principal Components Analysis, which is introduced to remove the effect that the light source directions are not so widely dispersed. The entire approach is non-parametric, empirical in that no explicit assumptions are made about light source directions or surface reflectance. Results are demonstrated by the experiments for real images.

View full abstract

An arbitrary direction face image reconstruction method using real images is proposed as a basic technology for virtual space teleconferencing systems. This method combines the conventional model-based and image-based approaches to enable the process with high speed and high precision. The input human face position and direction is measured using image processing techniques, and re ected upon a three-dimensional face model. By mapping the texture of the camera input image directly onto the model, realistic human face images could be reconstructed. Also, by mapping the input images consecutively in sequence, changes in facial expression could be approximated. A fundamental experiment using a home-use video camera and a relatively low-speed workstation showed the efficiency of the proposed method.

View full abstract

Simple method for extraction of paralinguistic features from a face as a nonverbal communicative intend was developed. A novel emotion-feature discrimination approach based on implementation of the natural compression and extraction abilities of humans was proposed. It combines simplicity and high computational efficiency in emotional classification and facial image processing. The method itself is divided into five steps: Segmentation by thresholding, face extraction and ellipse fitting, facial modules extraction, measurements and decision. Two independent, parallel experiments were obtained: real face-real face and real face-cartoon face. The cartoon face based model was developed by a paper emotional impression prototyping from which the emotion vector dimensionality was decided. The real face model was developed using Likert rating scales. It was demonstrated that our method works in both cases of real face based model and cartoon based model. At the same time we expressed the emotion in human face efficiently by a very small amount of information. We also showed that our method is applicable for emotion extraction from still and dynamic real face image.

View full abstract

Developing quality content may be the most difficult and costly task of developing a computer-based courseware. Despite this, little has been done to support the reuse of this material. In this paper we propose a particular way of structuring a courseware to conform a lesson unit in order to facilitate the use of the unit in different situations as well as the use of the material in different lessons. This is based on structuring a lesson’s computer-based learning material as a “didactic network”. A didactic network is a semantic network which represents the lesson unit. Although this approach was not developed exclusively having the wireless and mobile scenario in mind, it is a good example about developing courseware which should be used in different situations, including the mobile scenario.

View full abstract

Compound Images are mixed documents containing textual content along with continuous-tone or multi-level pictures. They are a very common form of documents found in textbooks, magazines, brochures, web sites etc. Because of the extremely distinct statistical nature of textual and multi-level images, compound image compression often involves multiple compression systems and a region segmentation scheme, such as the mixed raster content (MRC) approach. However, the new JPEG 2000 standard desires to have a coding system that is capable of compressing both continuous-tone and bi-level images progressively with similar system resources, and producing high quality images at low bit-rate, for example, below 0.25 bpp. Thus, in this paper, we search another approach, coding compound images with adaptive wavelet transform constructed by the lifting scheme. Though the wavelet transforms prove to be a powerful analysis tool for continuous-tone images, and provide superior performance for progressive transmission, their performance for compound and bi-level images is still not acceptable. The proposed adaptive discrete wavelet transform (ADWT), however, is able to reduce the entropy of bi-level region coefficients greatly while achieve nearly the same efficiency as the conventional transform for multi-level regions, by changing the wavelet basis adaptively according to the contents and embedding some non-linear processing to the transform. When evaluated by the JPEG 2000 verification model software version 0.0 (VM 0), the ADWT attains superior lossless compression ratio for compound and facsimile images versus conventional wavelet transforms, moreover, textual parts in a compound image are perfectly reconstructed at low bit-rate so that image quality of a compound image is improved at the early stage of transmission.

View full abstract