Norio Akamatsu’s research while affiliated with The University of Tokushima and other places

In this paper, a new image restoration approach using the Akamatsu transform is presented. In the method, by using repeated simple calculations based on the Akamatsu transform, an out-of-focus blurred image can be restored and sharpened. Out of focus is a major problem bothering many people in photography, especially the amateurs. There exits some solution to this problem on both the hardware and software sides. However, none offers a perfect solution. The Akamatsu transform can easily be embedded into hardware to offer faster processing due to its simplicity. Although, initially proposed for speech processing, this paper show the effectiveness of the transform in image processing. The Akamatsu transform is a combination of integral and differential transforms. The algorithm can be effective tools for image restoration in realtime image processing.

In this paper, we propose a fully automatic real time one image face gesture simulation using image morphing. Given a single image of a subject, we create several facial expressions of the face by morphing the image based on prior information stored in a data bank. The process involves the automatic detection of the control points both on the target image and the source data. The source data is a string of frames containing the desired facial expressions. A face detection neural network and a lips contour detector using edges and the SNAKES algorithm are employed to detect the face position and features. Five control points and the lips contour, for both the source and target images, are then extracted based on the facial features. Triangulation method is then used to match and warp the source image to the target image using the control points. In this experiment, using one expressionless face portrait, we create an animation to make it appear like the subject is pronouncing the five Japanese vowels. The final results shows the effectiveness of our method.

In this paper, we propose a system for automatic detection and recognition of vehicle license plates using template matching, genetic algorithms and neural networks. In the license plate detection part, an artificial template that consists of an outer rectangle that encloses two smaller rectangles is constructed. The size of the outer and inner rectangles, the positioning, size, orientation, position and color of the template are all controlled using the genetic algorithm. To fill the areas outside the inner rectangle, random selection of color pixels from a plate color database are used. In the character recognition part, we combine two methods into a hybrid system. In the first method, we train neural networks to recognize the characters and in the second method, we use template matching. To control the size of both the neural network inputs and the template, we apply a genetic algorithm to guide the search. The final system accuracy achieved is 96.8% for character recognition and 98.01% for license plate detection.

In this paper, we propose a classification system for fingerprint images that is based on the number of registered fingerprint persons. Most automated fingerprint identification systems use prior classification of fingerprint for improvement of efficiency verification using minutiae as features. However, methods that use fingerprint minutiae needs improvement because they are limited to the number of classable data. Therefore, many fingerprints are classified together, consequently taking a long time to match and verify a given fingerprint. In this work, we propose a system that classifies fingerprint patterns into individual classes. Instead of the classification using minutiae, we propose a classification system that is based on individual features and the number of registered persons. Efficiency verification improves because we donpsilat need to compare an input fingerprint image to all registered fingerprint images using this system. The proposed system carries out classification using neural network.

This paper presents a method for gender and age estimation which is robust to changing facial pose. We propose a feature point detection method, called the adapted retinal sampling method (ARSM), and a feature extraction method. A basic concept of the ARSM is to add knowledge about the facial structure to the retinal sampling method. In this method, feature points are detected on the basis of seven points corresponding to facial organs from a facial image. The reason why we used seven points as the basis of feature point detection is that facial organs are conspicuous in the facial region, and are comparatively easy to extract. As features robust to changing facial pose, skin texture, hue, and the Gabor jet are used for gender and age estimation. For classification of gender and estimation of age, we use a multilayered neural network. We also examine the left– right symmetry of faces in connection with gender and age estimation by the proposed method. © 2008 Wiley Periodicals, Inc. Electron Comm Jpn, 91(7): 32– 40, 2008; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecj.10125

Fast and automatic face detection from visual scenes is a vital preprocessing step in many face applications like recognition, authentication, analysis, etc. While detection of a single face can be accomplished with good accuracy, multiple faces detection in real time is more challenging not only because of different face sizes and orientations, but also due to limits of the processing power available. In this paper, we propose a real time multiple face detection method using multiple neural networks and an adaptive search area control method base on genetic algorithms. Although, neural networks and genetic algorithms may not be suitable for real time application because of their long processing times, we show that high detection accuracies and fast speeds can be achieved using small sized effective neural networks and a genetic algorithm with a small population size that requires few generations to converge. The proposed method subdivides the face into several small regions, each connected to an individual neural network. The subdivision guarantees small size networks and presents the ability to learn different face regions features using region-specialized input coding methods. The genetic algorithm is used during the real time search to extract possible face samples from face candidates. The fitness of the face samples is calculated using the neural networks. In the successive frames, the search area is adaptively controlled based on the information inherited from the proceeding frames. To prove the effectiveness of our approach we performed real time simulation using an inexpensive USB camera.

In this paper, a real time face detection method using several small size neural networks and a genetic algorithm with adaptive search area control is proposed. Neural networks and genetic algorithms may not be suitable for real time application because of their long processing times. However, in this paper, we show how fast speeds can be achieved using small effective neural networks and a genetic algorithm with a small population size that requires few generations to converge. We subdivide the face into several regions, each connected to an individual neural network. This guarantees small size networks and also offers the ability to learn different face regions features using different coding methods. The genetic algorithm is used during the real time search. It extracts possible faces from face candidates that are then tested using the neural networks. The face candidate area is then adaptively reduced depending on the location of the top six face samples. We then performed real time simulation using an inexpensive USB camera to prove the effectiveness of our proposal. We achieved between 98 and 96% accuracy for one or multiple faces respectively at 15 to 8 frames per second.

Previous studies of image restoration for noise images were based on a mask processing. These conventional noise removal methods based on the mask processing have an issue of defining degradation to accompany a spacial processing. In this paper, we propose a graininess suppression method based on edge shape. In this method, we detect edges from an noise image and perform graininess suppression for this image based on edge information. On the edge detection, we execute an image transformation for an image that enables us to extract edge by making principal component images. Moreover, we use the canny edge detection operator with can detect a weak edge that relates to a real edge, and do not detect a lie edge. In the suppression process, we use Wiener filter that can restore an noise image without making a complete edge map and the original signal map. We demonstrated the effectiveness of the present method for the noise added images and confirmed it by means of computer simulation.

Previous studies of image restoration for noise image were based on mask processing. These conventional noise removal methods represented from mask processing have issue of definition degradation to accompany spacial processing. In this paper, we propose a graininess suppression method based on edge shape. In this method, we detect edges from a noise image and perform graininess suppression for this image based on edge information. On the edge detection, we execute an image transformation for an image that enables us to extract edge by making principal component image. Moreover, we use the canny edge detection operator that can detect a weak edge that relates to a real edge, and do not detect a lie edge. In the suppression process, we use Wiener filter that can restore an noise image without making a complete edge map and the original signal map. We have that the present method for the noise added images to verify effectiveness and have confirmed this.

In this paper, a genetic algorithm guided control point extraction method that enables automatic face image morphing is proposed. This method can morph two face images by automatically detecting all the control points necessary to perform the morph. A face detection neural network, edge detection and medium filters are employed to detect the face position and features. Five control points, for both the source and target images, are then extracted based on the facial features. A triangulation method is then used to match and warp the source to the target image using the control points. Finally, color interpolation is done using a color Gaussian model that calculates the color for each particular frame depending on the number of frames used. A real coded Genetic Algorithm (GA) is used to aid the facial features extraction by overcoming size and orientation changes. We achieve very high quality morphs at high speed.