- big multimedia data analysis and user navigation so that ordinary users can generate high-quality media data that they really intended
- object/action recognition and machine learning
- optimization and high-speed processing
- 3D image/video processing
1. Big Multimedia Data Analysis and User Navigation
It is hard for ordinary users to make high-quality multimedia content such as good photos/videos, good annotations, etc. By mining big multimedia data on the internet, we are trying to develop systems that can navigate users to generate better media or suggest where/when/why to go, what/how to do to have better user experience. We are interested in processing not only images/videos but also voice/sound, text annotations, and documents. Some representative topics are
- Travel planning (route recommendation, photo shooting navigation, etc.),
- Social popularity estimation for SNS photos/videos and recommendation to enhance the popularity,
- Oral presentation mining.
2. Object/action Recognition and Machine Learning
We have been working on image recognition and machine learning algorithms that autonomously improves and fix its errors. We have been developing a confidence measure for machine learning algorithms. By our confidence analysis, we can tell with high accuracy whether the output from the machine learning algorithm is correct or not without knowing the ground truth. When the confidence is low, we can change the feature extraction and machine learning algorithms, thus achieving better recognition accuracy. In addition, we have been investing automatic parameter and architecture tuning by employing genetic algorithms and genetic programming.
3. Optimization and High-Speed Processing
With the increasing amount of multimedia data (for instance, we use more than 100 million photos), optimization and processing speed acceleration is becoming more critical. For this purpose, we have been working on coarse-to-fine processing, fast image processing with associative architecture, and optimization using GA/GP. There are many applications such as depth estimation using stereo matching, object tracking, seam carving of images/videos, image contrast enhancement, object recognition, and so on.
4. Other Research Topics
Our research topics are limited to the above. For instance, we are currently working on a prediction model on 3D model compression and its printed quality, real-time 3D face model reconstruction and synthesis using an RGB-D camera, recognition and classification of make-up photos, and so on.
- mail: yamasaki (at) hal.t.u-tokyo.ac.jp
- このサイトはCSS Designplateのテンプレートを使用しています