Method and system for generating and implementing orientational filters for real-time computer vision applications

• US 6,009,447 A
• Filed: 02/13/1997
• Issued: 12/28/1999
• Est. Priority Date: 02/16/1996
• Status: Expired due to Fees

• Current Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Original Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Inventors: Kubota, Toshiro, Alford, Cecil O.
• Primary Examiner(s): Palys, Joseph E.
• Assistant Examiner(s): NGUYEN, NGUYEN X
• Application Number:
  US08/801,602
• Time in Patent Office: 1,048 Days
• Field of Search: 364/724.1, 364/724.011, 364/724.17, 364/724.19, 364/572, 375/346, 382/275, 382/266, 340/706, 708/313, 708/308
• US Class Current: 708/313
• CPC Class Codes: G06T 5/20 using local operators

Abstract

Methods for generating and implementing digital orientational filters. A first method of the present invention is provided for generating digital orientational filters of uniform size but each having a different fixed orientation. A second method provides for dilation of the digital orientation filters generated by the first method, both in a decimated and an undecimated format. A third method of the present invention provides for steering the orientation of the filters generated by the first method. Also, associated VLSI hardware based systems implementing the above methods are disclosed. The above methods and systems allow digital orientational filters to be utilized in computer vision and other applications requiring a large amount of video signal data to be processed in real-time.

Claims

1. A method of processing video data, comprising the steps of:

A. providing a processor for processing input pixel data;

B. inputting filter parameters into said processor;

C. computing a set of 2D filter coefficients having a specific orientation from said filter parameters;

D. generating a single digital filter approximation matrix based on said computed filter coefficients; and

E. decomposing said approximation matrix into a plurality of separable digital filters by utilizing a set of orthogonal filter coefficients for each of said plurality of filters that provides an optimal approximation of each of said plurality of filters; and

F. increasing a dilation factor of said digital filters by a predetermined factor in order to generate an undecimated multi-resolution image decomposition output.

2. The method of claim 1, wherein said dilation factor comprises a 2×

2 dilation factor.

3. The method of claim 1, further comprising the steps of:

G. filtering pixel data;

H. outputting said filtered pixel data;

I. adjusting a dilation factor of said plurality of digital filters by a predetermined factor; and

J. repeating steps A-I for N levels of decomposition, where N is a positive, nonzero integer.

4. The method of claim 1, further comprising the steps of:

G. formulating a basic spline function to obtain pre-filter and low-pass filter coefficients; and

H. approximating a set of 1D high-pass filters by decomposing said basic spline function into separable functions.

5. The method of claim 1, further comprising the step of increasing a decimation factor of said filtered data in conjunction with said step of increasing a dilation factor of the orientation filters to provide a decimated multi-resolution decomposition filter output.

6. The method of claim 1, wherein said step of increasing a dilation factor comprises:

generating two-dimensional mother wavelet functions;

decomposing the mother wavelet functions into separable one-dimensional functions;

generating a basic spline function;

generating a low-pass filter based on a spline order, associated with said basic spline order function; and

approximating each of the separable one-dimensional filters in terms of said basic spline function to generate said plurality of digital filters.

7. The method of claim 6, wherein said step of approximating each of said separable one-dimensional functions comprises approximating each of said separable one-dimensional functions to generate interpolation, low-pass filters, and high-pass filters for recursive multi-resolution decomposition.

8. The method of claim 6, wherein said step of generating a basic spline function comprises the step of generating high pass filters in a horizontal orientation using the basic spline function given by the equations ##EQU23## and generating high-pass filters in the vertical direction using ##EQU24## where said above functions (a_i, b_i) are derived from a mother function of dyadic wavelets decomposed into a separable fashion given by the equation:

##EQU25##

9. The method of claim 1, wherein said step of decomposing comprises decomposing said approximation matrix into a plurality of sets of horizontal filters and at least one vertical filter that are implemented in a vertical-horizontal filtering scheme.

10. The method of claim 9, further comprising the step of inputting filtered data into a reorder buffer after the step of filtering the data through a vertical filter to minimize shift delay between the vertical and horizontal filters and to produce an undecimated filtered data output.