Minako Sawaki

Messages embedded in gaze of interface agents --- impression management with agent's gaze

We propose a gaze movement model that enables an embodied interface agent to convey different impressions to users. Managing ones own impression to influence the behaviors of others plays an important role in human communications. To create a new application area which involves agents in this kind of social interaction, interface agents that manage their impressions are required. For this purpose, we build the gaze movement model based on three gaze parameters picked from a large number of psychological studies: amount of gaze, mean duration of gaze, and gaze points while averted. In this paper, we introduce the gaze movement model and gaze parameters. We then present an experiment in which subjects evaluated the impressions created by nine gaze patterns produced by altering the gaze parameters. The results indicate that reproducible relations exist between the gaze parameters and impressions, which shows the validity of the model

Text-line extraction and character recognition of document headlines with graphical designs using complementary similarity measure

A method for recognizing characters on graphical designs is proposed. A new projection feature that separates text-line regions from backgrounds, and adaptive thresholding in displacement matching are introduced. Experimental results for newspaper headlines with graphical designs show a recognition rate of 97.7 percent.

Robust recognition of degraded machine-printed characters using complementary similarity measure and error-correction learning

Most conventional methods in character recognition extract geometrical features such as stroke direction, connectivity of strokes, etc., and compare them with reference patterns in a stored dictionary. Unfortunately, geometrical features are easily degraded by blurs, stains and the graphical background designs used in Japanese newspaper headlines. This noise must be removed before recognition commences, but no preprocessing method is completely accurate. This paper proposes a method for recognizing degraded characters and characters printed on graphical background designs. This method is based on the binary image feature method and uses binary images as features. A new similarity measure, called the complementary similarity measure, is used as a discriminant function. It compares the similarity and dissimilarity of binary patterns with reference dictionary patterns. Experiments are conducted using the standard character database ETL-2 which consists of machine-printed Kanji, Hiragana, Katakana, alphanumeric, an special characters. The results show that this method is much more robust against noise than the conventional geometrical feature method. It also achieves high recognition rates of over 92% for characters with textured foregrounds, over 98% for characters with textured backgrounds, over 98% for outline fonts, and over 99% for reverse contrast characters.

Robust feature extraction based on run-length compensation for degraded handwritten character recognition

Conventional features are robust for recognizing either deformed or degraded characters. This paper proposes a feature extraction method that is robust for both of them. Run-length compensation is introduced for extracting approximate directional run-lengths of strokes from degraded handwritten characters. This technique is applied to the conventional feature vector based on directional run-lengths. Experiments for handwritten characters with additive or subtractive noise show that the proposed feature is superior to conventional ones over a wide range of the degree of noise.

Character recognition in bookshelf images by automatic template selection

Presents a multiple-dictionary method for recognizing low-quality characters in scene images. First, the environmental conditions of an input image are estimated using an initial dictionary. Then, a relevant dictionary from multiple dictionaries reflecting different environmental conditions is automatically selected from the estimation and used for recognition. Experiments are made for characters in images of bookshelves. The results show that the proposed method achieves a higher recognition rate (89.8%) than that obtained by using a single dictionary (76.4%). Furthermore, recognition accuracy improves from 89.8% to 95.2% using contextual postprocessing.

Text-line extraction and character recognition of Japanese newspaper headlines with graphical designs

The conventional OCR fails to recognize most characters in Japanese newspaper headlines with graphical designs because of the difficulty of removing the designs. This paper proposes a method that recognizes such characters without removing the designs. First, text-line regions are extracted from a local distribution of the combination of black and white runs observed in a rectangular window while the window is shifted pixel-by-pixel in the direction of the text-line. Characters in the extracted text-line region are then recognized by displacement matching. Adaptive thresholding against the degree of degradation suppresses spurious candidates yielded by displacement matching even with graphical designs. Experimental results for fifty Japanese newspaper headlines show that the method achieves a recognition rate of 97.7%, much higher than a conventional method (17.0%).

Robust character recognition using adaptive feature extraction

This paper describes an adaptive feature extraction method that exploits category specific information to overcome both image degradation and deformation. When recognizing multiple fonts, geometric features such as directional information of strokes are often used but they are weak against the deformation and degradation that appear in videos and natural scenes. To tackle these problems, the proposed method estimates the degree of deformation and degradation of an input pattern by comparing the input pattern and the template of each category as category specific information. This estimation enables us to compensate the aspect ratio associated with shape and the degradation in feature values. Recognition experiments using characters extracted from videos show that the proposed method is superior to the conventional alternatives in resisting deformation and degradation.

Character recognition in bookshelf images using context-based image templates

This paper proposes a method for recognizing degraded characters in bookshelf images captured by a digital camera. We adopt displacement matching and templates that include neighboring characters or parts thereof to cope with the degradation. The templates are referred as context-based image templates, since they offer more contextual information than single-letter templates. Such templates are effective wherever there is a restricted word set, such as journal titles, year, month, volume, and number. Experiments with 3,468 characters in nine bookshelf images show that this method achieves a higher recognition rare (96.3%) than single-letter templates (88.4%).

Lighting Independent Skin Tone Detection Using Neural Networks

Skin tone detection in conditions where illuminate intensity and/or chromaticity can vary often comes with high computational time or low accuracy. Here a technique is presented integrating chromaticity and intensity normalization combined with a neural skin tone classification network to achieve robust classification faster than other approaches.

Category-dependent feature extraction for recognition of degraded handwritten characters

Conventional methods for recognizing multiple fonts and handwriting are generally robust against deformation, but are weak against degradation. This paper proposes a category-dependent feature extraction method that resists both deformation and degradation. Our proposed method compares an input pattern with the template of each category and estimates the degree of degradation of the input pattern. Approximate stroke run-lengths without degradation are then obtained by compensating the inaccurate runs caused by degradation. Recognition experiments using degraded handwritten characters show that the proposed feature is superior to conventional ones in resisting degradation.