Yoichi Kageyama

Generation of emotional feature space for facial expression recognition using self-mapping

This paper proposes a method for generating a subject-specific emotional feature space that expresses the correspondence between the changes in facial expression patterns and the degree of emotions. The feature space is generated using self-organizing maps and counter propagation networks. The training data input method and the number of dimensions of the CPN mapping space are investigated. The results clearly show that the input ratio of the training data should be constant for every emotion category and the number of dimensions of the CPN mapping space should be extended to effectively express a level of detailed emotion.

Analysis of the segments extracted by automated lineament detection

Lineaments are important features showing subsurface elements or structural weaknesses such as faults. Most lineament maps have been drawn based on fieldwork by experts and visual analysis of enhanced image data. In visual interpretation and mapping of lineaments, geologists use their knowledge and experience to extract the lineaments from the curved and straight lines in image data. A different expert may extract different segment elements through a visual approach. In order to conduct the lineament detection under the same conditions, automatic extraction for lineaments though image processing is useful. In an earlier paper, an extraction method for lineaments using airborne Synthetic Aperture Radar (SAR) data was presented. The results indicated that the segments given by the method agree with a lineament map drawn by experts. The objective of this paper is to examine the relationship between extracted segments from both airborne SAR and Landsat 5 thematic mapper (TM) data and geographical features. River erosion creates various stream patterns that are influenced by the subsurface structure. Many lineaments can be extracted from a drainage system for topography. This paper seeks to compare extracted segments and water systems. Also, relief energies can indicate the level of the river erosion. The relief energy at a study site has been compared and the correspondence between the computation and extracted segments is described. Finally, the difference between segments in SAR and TM data has been compared.

Background replacement using chromatic adaptation transform for visual communication

Chromatic adaptation, an ability of human vision, was implemented in video chat.Illuminant colors of landscapes are required by the function of chromatic adaptation.We estimated landscape illuminant colors based on the dichromatic reflection model.Background replacement using chromatic adaptation produces the sense of realism. Replacing a video chat background with a landscape image can generate the realism of a user actually being in the landscape. To enhance this realism, we proposed in our previous study a background replacement method that uses a chromatic adaptation transform. This method can enhance the realism of video chat by fitting the color of the foreground image to an illuminant color of a landscape, which is used as the new background image. However, if an incorrect color of the landscape illuminant is obtained through this method, which estimates the illuminant color on the basis of a gray world assumption, the method might not enhance the realism. This is because it converts the foreground color to an incorrect color. In this paper, we therefore propose a method to estimate illuminant color on the basis of the dichromatic reflection model, which improves background replacement using the chromatic adaptation transform. We perform a subjective evaluation using 13 subjects to examine the effects of the proposed method. The results indicate that the proposed method can effectively enhance the realism of the background replacement video.

Use of Dichromatic Reflection Model to Estimate the Illuminants of Landscape Images for Background Replacement

Background replacement using a landscape image can generate realism when video chatting, making the user feel as if they are actually at the location shown in the landscape. In our previous study, to enhance this realism, we proposed a background replacement method using a chromatic adaptation transform. This method fits the color of a foreground image to the illuminant of the landscape used as the new background image, and can enhance the realism of the video chat. However, if an incorrect illuminant of the landscape is obtained based on gray world assumption, the method might not enhance realism because it fits the foreground color to this illuminant. In this paper, we propose an illuminant estimation method that uses the dichromatic reflection model, and describe the appropriate parameters for this method. Our experimental results suggest that landscape illuminants can be estimated by the proposed method better than a gray world assumption.

Automatic land cover classification at sub-pixel scales by using NOAA AVHRR data

This study proposes an automatic land cover clas- sification algorithm at sub-pixel scales by using the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) data. It is well known that land cover conditions in the NOAA AVHRR data are classified into three different classes: sea, land, and cloud. Super- vised data of the three classes are automatically sampled from the features of a histogram obtained from multispectral bands. With the supervised data, pure pixels for each class are classified by fuzzy reasoning. Subsequently, the remaining pixels are classi- fied into three types of mixed pixels (mixels) by using fuzzy rea- soning and edge information. The classification results by the proposed method are in considerable agreement with the actual land cover conditions. This study also suggests that the proposed approach provides reasonable results when compared with the maximum likelihood method and k-means clustering.

A study of learning data size for automatic face area detection in sequential thermal images

Chronological change of temperature on cheeks includes important information to detect an emotion occurrence. To measure the specific region of face skin temperature accurately, we have developed a face detection method from sequential thermal image acquired in 30fps. In this paper, we investigated minimum quantity of learning data that is sufficient to create a high accurate face area detector. The experimental results for five persons showed that high detection rate was obtained when using over 350 images.

Relationship between physical conditions and lip motion change arising due to amusement feeling

In our previous study, we have developed a novel algorithm for lip motion analysis arising due to amusement feeling after watching TV-comedy programs as emotion-eliciting stimuli. This algorithm can detect the occurrence of amusement feeling. However, relationship between physical conditions of subjects and the variance of lip motion features extracted from pronouncing some sentences have not yet been analyzed. This paper analyzes this relationship and the results of the conducted experiment show that the variance of lip motion features varies due to physical conditions of subjects.

Quantification of Emotions for Facial Expression: Generation of Emotional Feature Space Using Self-Mapping

The shape (static diversity) and motion (dynamic diversity) of facial components, such as the eyebrows, eyes, nose, and mouth, manifest expression. From the viewpoint of static di‐ versity, owing to the individual variation in facial configurations, it is presumed that a facial expression pattern due to the manifestation of a facial expression includes subject-specific features. In addition, from the viewpoint of dynamic diversity, because the dynamic changes in facial expressions originate from subject-specific facial expression patterns, it is presumed that the displacement vector of facial components has subject-specific features.

Feature analysis of groundwater discharge points in coastal regions around Mt. Chokaisan, Japan by using alos palsar data

Submarine groundwater discharges exist in the Japan Sea around Mt. Chokaisan, Japan. However, in the coastal regions, the details regarding their properties have not yet been clarified. In our previous study, we have detected groundwater discharge points arising due to the difference in freshwater and seawater by using the Landsat ETM+ and the ALOS AVNIR-2 signals. The ETM+ and AVNIR-2 are passive sensors, and the data taken by the above sensors will be affected by clouds, the limited data are available. In order to understand features of the groundwater discharge points regardless of the weather conditions, this paper analyzes textures obtained from the ALOS PALSAR data. Our experimental results were in considerable agreement with the results obtained from the AVNIR-2 data, and the realities in the study area.

Algorithm for Extracting Ground Control Points from NOAA-AVHRR Data

National Oceanic and Atmospheric Administration (NOAA) and Advanced Very High Resolution Radiometer (AVHRR) data are available on a daily basis and have been frequently used for global observation. Geometric correction is very important as a preprocessing technique for long-term analysis or monitoring that makes use of the data. Ground control points (GCPs) are set to manually determine the geometric transform of the equation because 4-10 pixels of error can arise in the geometric correction based on only the system information. Therefore, we propose an algorithm that automatically extracts GCPs with information of run-length matrix and on adjacent areas. The proposed method has two steps: the creation of measurement images and measurements of the run-length matrix. Results suggest that the proposed approach is suitable for extracting GCPs from NOAA-AVHRR data.