Ryota Mase

BRIGHT: A scalable and compact binary descriptor for low-latency and high accuracy object identification

This paper proposes a new scalable and compact binary local descriptor, named the BRIGHT (Binary ResIzable Gradient HisTogram) descriptor, for low-latency and high accuracy identification of real-world objects in images. The BRIGHT descriptor is extracted by first creating a hierarchical HOG (Histogram of Oriented Gradients) of a local patch centered around keypoints detected from an image. The elements of the histogram are then binarized, and the subset of bits is progressively selected forming a progressively scalable descriptor with a size ranging from only 32 bits to 150 bits. Experiment using images with objects taken under various camera viewpoints, lighting conditions, and occlusions, shows that the BRIGHT descriptor can robustly match objects with an identification accuracy comparable with that of SIFT descriptor, but at a descriptor size smaller than 1/10 of SIFT. With the reduced descriptor size, transmission of descriptors from a mobile device to a database server can be dramatically speeded up, enabling low-latency response in mobile search services.

Background scene classification robust to the influence of human regions

We propose a background scene classification method robust to the influence of human regions. Conventional methods classify scene of an image by using image features extracted from entire region in the image. Therefore, in these methods, the influence of the human region such as color of the skin and the clothes reduces classification accuracy of the background scene. Our method classifies background scene of an image by using image features extracted from only background region except detected human regions. The experimental results show that the proposed method improves average of the rate at the balance point between recall rate and precision rate in almost all background scenes compared to the conventional method.

Credit-title detection of video contents based on estimation of superimposed region using character density distribution

We propose a credit-title detection method of video contents based on estimation of superimposed region using character density distribution. Copyright information of video contents is manually extracted for the secondary use of those contents, and its cost is highly expensive. Therefore, automatic detection of credit titles that contain copyright information is highly demanded. However, accuracy of conventional methods is usually insufficient for this purpose. Our method first estimates credit-title-superimposed region based on character density distribution calculated in advance by using many video contents. Then, credit titles are detected in the estimated region. The experiment results show that proposed method improves both recall and precision rates compared to a conventional method. Furthermore, the processing time of the proposed method is less than half that of the conventional method for all contents.