Related Experiment Videos
Improved structures of maximally decimated directional filter banks for spatial image analysis
Sang-Il Park1, Mark J T Smith, Russell M Mersereau
1Motorola, Inc, Austin, TX 78729, USA. sang.park@motorola.com
Summary
This study presents an improved directional filter bank (DFB) structure that preserves visual information and offers efficient implementation. The new design reduces complexity significantly compared to Quadrature Mirror Filters (QMFs).
Area of Science:
- Signal Processing
- Image Analysis
- Filter Design
Background:
- Directional Filter Banks (DFBs) are crucial for subband decomposition in image processing.
- Existing DFB structures can introduce visual distortions and lack optimal efficiency.
- Quadrature Mirror Filters (QMFs) offer a baseline for comparison but have limitations.
Purpose of the Study:
- To introduce an improved DFB structure that preserves visual information in the subband domain.
- To maintain efficient polyphase implementation and exact reconstruction properties.
- To reduce computational complexity compared to existing filter banks.
Main Methods:
- Developed a novel DFB structure with post-sampling matrices to eliminate visual distortions.
- Outlined a systematic framework for analyzing and improving DFBs.
- Designed efficient linear phase Infinite Impulse Response (IIR) prototype filters.
Main Results:
- The proposed DFB structure successfully preserves visual information.
- Achieved exact reconstruction and efficient polyphase implementation.
- Demonstrated over a three-fold complexity reduction compared to QMFs using linear phase IIR filters.
Conclusions:
- The improved DFB structure offers significant advantages in visual information preservation and implementation efficiency.
- The use of linear phase IIR filters is key to achieving substantial complexity reduction.
- This work provides a valuable advancement for subband image processing applications.