This paper presents a new approach for recognizing face images having much lower quality in terms of low resolution and contrast than that of the training face images. The input images also contain many faces having multi poses variations because the images were captured under different lighting, sensor, and camera angle. The difficulties of this system are not only in face detection process but also in the recognition process. To handle the difficulties, the improvement of Haar-Like features based face and the Fusion of Scale Invariant Features (FSIF) based classifier are employed, respectively. In addition, the FSIF is integrated with the Incremental Predictive Linear Discriminant Analysis (IPDLDA) for decreasing the time computation of direct FSIF based classification technique. The proposed method can overcome the mentioned difficulties which is shown by high enough recognition rate compared to stand alone FSIF and SIFT-based methods with less than 2 second recognition time.
Effect of human-body swing on visibility of scrolled texts with direction dependency is tested two ways, a subjective test and eye movement measurements. The results show that (1) recognition rates of scrolled characters are decreased/increased compared with the rates in body-stationary conditions when characterscrolling direction is congruent/incongruent with body-yawing direction, and (2) in most cases, observed eye movement patterns were similar regardless of the character recognition rate. The important findings are that (1) apparent eye movement pattern is affected by body swing direction because of the vestibulo-ocular reflex (VOR), but (2) the net eye movement after subtracting or compensating the effect of VOR is not affected by body swing direction. These results suggest a higher stage of visual system processing than reflection is affecting the visibility, which is modulated by the relationship between stimulus-motion and body-rotation direction.
There are not so many pixels on the numbers of car license plate which is obtained by surveillance camera, so the numbers can not be discriminated without endeavor. In this paper, we propose the novel method which discriminate the numbers of car license plate on low resolution by employing the moment characteristics ( skewness , flatness, aspect ratio ) for Bayesian discrimination method. The target number on low resolution is assumed to be composed of not so many pixels, e.g. four pixels on vertical direction, and two pixels on horizontal direction. We analyzed the effectiveness of our proposed method by the computer simulation for the learning data which is generated by low resolution model for car license plate numbers. The discriminated results of the actual car license plate numbers on low resolution show the usefulness of our proposed method.
Cheyne-Stokes respiration (CSR) has a high prevalence among newborns, especially among preterm babies. Although doctors generally recognize the phenomenon, they are not able to assess the severity of CSR in individual infants. CSR is characterized by cyclical weakening and strengthening of respirations with apnea. In this study, we developed an apnea detection method, and a CSR detection method using detected apneic events. We detected apnea using two features of respiratory waveforms. The first feature is frequency information calculated from wavelet coefficients. The second is information based on the shape of the waveform. In our CSR detection method, we used a spurious periodicity feature to determine CSR sections. The waveform is calculated by a respiratory monitoring system that uses a fiber-grating vision sensor to measure the vertical motion of the infant's thoracic and abdominal regions during respiration. Our method is effective at detecting apnea (sensitivity: 94.3 %, specificity: 99.7 %).
We propose a new textured 3D reconstruction method that uses image sequences output by projection-based display systems. So that image sequences can recreate convex objects on a planar display, we extend the photometric model used in photometric stereo to handle full color images. According to the photometric model, we generate multiple shading images from surface normals on the original 3D object with varying light directions. We show that an image data matrix created from the multiple images can be decomposed into a surface normal matrix scaled up to the color intensity of three channels and a light-source directional matrix, and that its rank is at most three. In order to ensure robust and unique matrix decomposition, we introduce new metric constraints by using pairs of the directional lights that correspond to odd and even frames. Based on matrix decomposition, we describe a method that reconstructs the surface normals for 3D shape and surface texture. We examine the proposed method by challenging it with both computer-simulated data and real images. These results demonstrate that our method yields good 3D reconstruction and surface texture.
To generate realistic representation of the nature scene is one of the most challenging areas in the computer graphics community. Ray tracing is the most well-known technique to synthesize a realistic image. Since ray tracing is the most suitable method for simulating reflection and refraction of the light, it has been used for simulating various atmospheric optical phenomena caused by them. Mirage is a kind of atmospheric optical phenomena. Therefore, it is possible to replicate mirages in 3DCG by simulating or modeling condition of the air. We focus on pencil tracing technique that is an extension of the conventional ray tracing technique based on the paraxial approximation theory. Our simple method based on pencil tracing can efficiently generate the appearance of mirages without complex thermodynamic simulation.
We propose a method for generating pointillistic style images from input image considering the features of Seurat's pointillism. Georges-Pierre Seurat is a pioneer and prime exponent of neo-impressionist. He established a technique called pointillism based on scientific color theory. There are three important features of Seurat's pointillism: optical mixture, complementary color contrast and halo effect. The most important feature is the optical mixture. To implement the optical mixture faithfully, we present a pointillistic halftoning method for color halftoning on random dots by utilizing a spatial data structure of boundary sampling algorithm. In addition, we implement complementary color contrast and halo effect according to actual Seurat's painting steps.
Interests in food safety have been growing and identification management has been recognized very important to improve qualities and safety of foods. This paper investigates pig face recognition to enable inexpensive marker-less pig identification management systems. Eigenspace methods are known to be effective to human face recognition, and we applied them to pig face recognition. From experimental results, we found that pig eyes are the most effective face part for recognition. We obtained 97.9% recognition rate for 16 categories with 16 samples/category training data, and some potential of the method was suggested.
This paper presents a new evaluation criterion for visualization of image search results based on the feature integration theory. This criterion is derived by combining two elements, visual saliency on visualization and grouping degree of similar images. Visual saliency, which is calculated from the feature integration theory, on visualization of image search results enables representation of users' attention, which is closely related to the effectiveness of finding images. Furthermore, since users perceive similar images that are close to each other as one group, grouping degree of similar images enables evaluation of the effectiveness when users find images similar to a desired image. Therefore, by combining visual saliency on visualization and grouping degree of similar images, we can derive the novel criterion and evaluate the effectiveness of visualization of image search results.
This paper proposes a new diminished reality technique which removes AR markers from a user's view image. In order to achieve natural marker hiding, three factors should be considered; (1) naturalness of texture generated on a marker area. (2) geometric consistency between consecutive frames, (3) photometric consistency between a marker area and its surrounding. In this study, assuming that an area around a marker is locally planar, the marker area in the first frame image is inpainted using the rectified image to achieve high-quality inpainting. The unique inpainted texture is overlaid on the marker region in subsequent frames according to camera pose for temporal geometric consistency. Global and local luminance changes around the marker are reflected to the inpainted texture for photometric consistency.
The LEGO brick system is one of the most popular toys in the world. It can stimulate one's creativity while being lots of fun. It is however very hard for the naive user to assemble complex models without instructions. In this work, we propose a method that converts 3D polygonal models into LEGO brick building instructions automatically. The most important part of the conversion is that the connectivity between the bricks should be assured. For this, we introduce a graph structure named "legograph" that allows us to generate physically sound models that do not fall apart by managing the connections between the bricks. We show some experimental results and evaluation results. These show that the 3D brick models generated following the instructions generated by our method do not fall apart and that one can learn how to efficiently build 3D structures from our instructions.