Two-dimensional filter bank design for optimal reconstruction using limited subband information.

Area of Science:

• Signal Processing
• Image Analysis
• Data Compression

Background:

• Perfect Reconstruction Filter Banks (PRFB) are crucial for signal decomposition and reconstruction.
• Existing PRFB designs may not be optimal for applications requiring reduced subband representations.
• Image compression and pattern recognition benefit from efficient low-resolution signal representations.

Purpose of the Study:

• To propose novel design techniques for 2-D PRFB analysis and synthesis filters.
• To achieve optimal reconstruction with a reduced number of subband signals.
• To adapt filters to input image statistics for energy concentration in specific subbands.

Main Methods:

• Minimization of squared error between original signal and low-resolution representation.
• Extension of frequency domain principal component analysis to two dimensions.
• Analysis of general 2-D discrete nonstationary and stationary second-order processes.

Main Results:

• Optimal 2-D filters are generally nonseparable.
• For separable random fields, only the first and last PRFB filters are separable.
• Concentration of signal energy in initial subbands is achieved, enhancing compression and classification.

Conclusions:

• The proposed optimal PRFB design techniques are effective for image compression and pattern representation.
• The developed filters enable efficient low-resolution analysis by concentrating signal energy.
• The nonseparable nature of optimal filters for general 2-D processes is demonstrated.