Automatic design of morphological algorithms for machine vision
View Patent ↗The present invention provides a technique for automated selection of a parameterized operator sequence to achieve a pattern classification task. A collection of labeled data patterns is input and statistical descriptions of the inputted labeled data patterns are then derived. Classifier performance for each of a plurality of candidate operator/parameter sequences is determined. The optimal classifier performance among the candidate classifier performances is then identified. Performance metric information, including, for example, the selected operator sequence/parameter combination, will be outputted. The operator sequences selected can be chosen from a default set of operators, or may be a user-defined set. The operator sequences may include any morphological operators, such as, erosion, dilation, closing, opening, close-open, and open-close.
1. A method for automated selection of a parameterized operator sequence to achieve a pattern classification task, comprising die steps of:
inputting a collection of labeled data patterns;
deriving statistical descriptions of the inputted labeled data patterns;
determining a criterion function which is used to derive classifier performances by performing the steps of:
determining a classifier performance for each of a plurality of candidate operator sequences and corresponding parameter values using the derived statistical descriptions by performing the steps of:
for each candidate operator sequence and corresponding parameter values, performing:
constructing an Embeddable Markov Chain (EMC), given the derived statistical descriptions for the input data patterns and output statistic to be calculated; and
calculating ouput statistics using the EMC, the output statistics a function of the derived statistical descriptions for the inputted data patterns and a Boolean transformation,
identifying an optimal classifier performance among the determined classifier performances according to specified criteria; and
selecting the operator sequence and corresponding parameter values, associated with the identified optimal classifier performance.
2. The method of claim 1 , further including the step of outputting performance metric information.
3. The method of claim 2 , wherein the outputted performance metric information includes the selected operator sequence and corresponding parameter values.
4. The method of claim 1 , wherein the step of deriving statistical descriptions for input patterns includes deriving the statistical descriptions using a probability model.
5. The method of claim 4 , wherein the probability model is a mixture of Hidden Markov Models (HMMs).
6. The method of claim 4 , wherein the probability model is a Bayesian network.
7. The method of claim 4 , wherein the probability model employs a non-parametric density representation.
8. The method of claim 1 , the inputted collection of data patterns includes at least one of patterns of interest and patterns of non-interest.
9. The method of claim 1 , wherein the plurality of operator sequences includes default operator sequences.
10. The method of claim 1 , wherein the plurality of operator sequences includes operator sequences defined by a user.
11. The method of claim 1 , wherein the specified criteria relates to maximum expected classifier performance.
12. The method of claim 11 , wherein the maximum expected classifier performance relates to balancing the tradeoff between false alarm errors and miss detection errors.
13. The method of claim 1 , wherein the specified criteria are defined by a user.
14. The method of claim 1 , wherein the plurality of operator sequences includes at least one of erosion, dilation, closing, opening, close-open, and open-close.
15. The method of claim 1 , wherein the plurality of operator sequences includes an operator that maps an input Boolean vector to an output Boolcan vector.
16. The method of claim 1 , wherein the plurality of operator sequences includes an operator that is defined as successive application of a 1D filter in two orthogonal directions.
17. The method of claim 1 , wherein the inputted collection of data patterns includes gray-level data transformed to a binary representation.
18. The method of claim 1 , wherein the inputted collection of data patterns includes color data transformed to a binary representation.
19. The method of claim 1 , wherein the step of constructing an EMC comprises the steps of:
(a) constructing a state space;
(b) building a state-transition graph with associated state-transition probabilities for the candidate operator sequence; and
(c) partitioning the state-space.
20. A method for determining optimal classifier performance of a plurality of candidate operator sequences and corresponding parameter values, comprising the steps of:
for each candidate operator sequence and corresponding parameter values, performing:
(a) constructing an Embeddable Markov Chain(EMC), given statistical descriptions for inputted data patterns and output statistic to be calculated;
(b) calculating the output statistics using the EMC; and
(c) selecting an optimal operator sequence and corresponding parameter values using the output statistics, according to specified criteria.
21. The method of claim 20 , wherein the step of constructing an EMC comprises the steps of:
a) constructing a state space;
b) building a state-transition graph with associated state-transition probabilities for the candidate operator sequence; and
c) partitioning the state-space.