Tektronix - Beaverton, Oregon Campus --- 6:00pm - 7:30pm
14150 SW Karl Braun Drive - Building 50
Detecting and scoring (MOS) video quality based on packet loss, or partial decoding of the stream to determine how much quantization was done at the encoder, can only score the last encode and is not sensitive to blur or other defects in the uncompressed source. Also, most No-Reference (NR) video quality analyzers that do decode and analyze each frame, search for signatures of specific defects such as filtering to detect the loss of detail, or blur or block edges due to tiling. The results of these measurements need to be pooled to create a useful NR MOS score. What is needed is a Machine Learning approach, that is trained on very large (big data) data base of images with various levels and types of distortion, pre-scored with established Full-Reference (FR) methods like MS-SSIM, that learns how to map a set of features into an estimated NR MOS score that matches the FR score. This way, no reference images are needed and a Blind, NR MOS scoring method can be used over a wide range and combinations of image distortions. Tonight's speaker, Dan Baker, is a key Tektronix Video Engineer, and longtime SMPTE and IEEE member. Dan has a B.S. and M.S. in Electrical Engineering, with current academic studies in Advanced Digital Signal Processing, and Machine Learning applied to Image Processing. He co-authored two papers published in the SMPTE Journal, and has over 40 issued patents in video, signal and image processing, with several pending. He is currently working on new HDR/WCG measurement methods, as well as a No-reference Video Quality measurement. Beverages provided. Parking is Free.