查看更多>>摘要：We introduce a framework on a general order pyramid transform for variational multiresolution registration. The traditional pyramid transform shrinks the image to an image of the quarter size. If the result of the pyramid transform is expressed in the same landscape with the original images, the result of the transform yields a low-resolution image. The pyramid transform is achieved by subsampling after linear smoothing. Furthermore, the dual operation of the pyramid transform is achieved by linear interpolation after upsampling. The rational-order pyramid transform is decomposed into subsampling by linear interpolation and the traditional pyramid transform. By controlling ration between subsampling for linear interpolation and subsampling in the pyramid transform, we construct the rational-order pyramid transform.
查看更多>>摘要：Gaze tracking is an important tool in many domains. Recent development in Convolutional Neural Networks (CNN) has allowed invention of gaze tracking techniques that work on commodity hardware such as a camera on a personal computer. Moreover, it has been shown that information from the full-face region can provide better performance than from an eye image alone. However, the problem with using the full-face image is the heavy computation cost due to the large image size. This study tackles this problem by efficiently compressing face images using importance weights to face regions. It is shown that the image compressed with the proposed method preserves the accuracy than the image resized as it is.
查看更多>>摘要：In deep learning, in order to improve learning performance, preprocessing and ingenuity to combine a plurality of discriminators are performed. It can be inferred that it has elements exceeding the set of learning. Therefore, a configuration to combine multiple recognition elements with low loss will be studied. The advance category classification method is expected to narrow the scope of learning in the next stage. Combining elements specialized for FalsePositive/FalseNegative removal after the positive/negative determination is considered to be effective if the accuracy of the subsequent stage is high. We conducted a license plate recognition experiment by combining these and achieved the best performance for Caltech data.
查看更多>>摘要：本稿では，並進·回転·スケール変動のある画像を高速に照合できるパターンマッチング手法を提案する．従来，バイオメトリクスや人工物メトリクスによる個体識別など，特に高精度な画像マッチングが求められる分野で，画像間の相関に基づくパターンマッチング手法が用いられている．一般的な手法では，撮影時に不可避な幾何変動を補正するために，相関演算を繰り返し行い，空間領域での相関値マップを求める．その処理量は多く，大規模なデータベースから個体を識別·照合する用途においては，処理速度の面で課題があった．本稿では，画像のFourier-Mellin 変換後の周波数特徴の正規化クロスパワースペクトルを求め，その周波数領域での分布形状を判別することで，照合する画像間の同一性を高速に判定する手法を提案する．実験では，提案手法を物体指紋認証による工業製品の個体識別に適用した．均一に塗装されたプレート部品11，571個体を誤りなく識別できる照合精度と，汎用デスクトップPCを用いて1 vs 11,571照合を約0.83秒で処理できる高速性を確認した．
查看更多>>摘要：This paper presents an experimental study on speaker verification performances using air and ear microphones in various acoustic conditions. Most existing speaker verification systems use an air microphone. Such systems often suffer from real environments which practically include background noise and/or reverberation. Ear microphone, whose transmission bandwidth is often limited due to skin or bone conduction, gives poor performance as compared with air microphone under ideal condition, while it is worn on the user's ear and hence robust to background noise. This paper attempts to discover suitable conditions for speaker verification systems using those two microphones. Effective combination of the microphones is also studied in terms of score fusion.
NETL
NSTL