Pilot Tests on Training Sequences

1
Pilot Tests on Training Sequences
Elastic Algorithms for Region of Interest Video
Compression, with Application to Mobile Telehealth
Sira Rao and Dr. Nikil Jayant Multimedia
Communications Lab School of Electrical and
Computer Engineering Georgia Institute of
Technology, Atlanta, GA
Encoder State Machine
Encoder State Table Region-Quality Mappings
Randomly Assigned Video Sample Number Bit Rate (kbps) Feature Sets Adequate to Assess? Clinical Assessment
Rn1 100 Rib Retraction Oral Cyanosis Tripod Posture 1-5 1-5 1-5 Comments Comments Comments
. . . . Rib Retraction Oral Cyanosis Tripod Posture 1-5 1-5 1-5 Comments Comments Comments
Rn10 1000 Rib Retraction Oral Cyanosis Tripod Posture 1-5 1-5 1-5 Comments Comments Comments
ML PL DL BE
ML 1 2 3 4
PL 5 6 7
DL 8 9
BE 10
Performance Results
Detection of Transitions
600kbps 0.40 28.6
600kbps 0.19 25.2
State Transitions due to Inadequate Quality

• Manual trigger of inadequate quality within the
  ROI
• Automatic trigger through a PSNR difference
  method
• Based on expectation of minimal difference in ROI
  and BKGRND quality
• Offsets underestimation in nominal mappings
• PSNR_TH state dependent i.e. expect higher
  differences where ROI and BKGRND have different
  quality abstractions (e.g. state 7 (PL ROI, BE
  BKGRND) and state 5 (PL ROI, PL BKGRND))

• Mapping from bits to quality levels may not
  always be satisfactorily achieved
• Different medical features with likely different
  compression limits on PL, DL
• Nominal mapping possibly the average value
• Thus, when ROI not of PL or DL quality when
  current state indicates it to be so, transition
  to a lower state in the same row to increase
  R_ROI

900kbps 0.56 29.5
900kbps 0.29 27.7
Quality Level ROI BKGRND
PL 0.3157 0.3157
DL 0.2104 0.2104
BE - 0.1052
BE_MIN - 0.0526
600kbps 0.19 18.0
600kbps 0.36 20.4
600 600 900 900 1200 1200
ROI BKGRND ROI BKGRND ROI BKGRND
Foreman 0.40 0.12 0.56 0.21 0.62 0.31
Flower 0.36 0.14 0.53 0.21 0.58 0.33
Football 0.38 0.13 0.55 0.21 0.62 0.31
MPEG2 0.19 0.19 0.29 0.29 0.39 0.39
900kbps 0.53 21.1
900kbps 0.29 18.8