Robust method for image feature estimation
View Patent ↗Noise or outliers corrupt image non-contact measurement of a geometric structure or geometric entity. A weight image is created prior to fitting whose pixel value indicates certainty of image information or feature signal strength. Learning images can enhance the weight image or it can be adjusted by iteration to achieve robust fitting results.
1. A robust method for image feature estimation comprising:
a. receiving at least one learning image input;
b. accumulating a weight image from the at least one learning image;
c. processing the at least one learning image using the accumulated weight image to produce a weight image output wherein the weight image is derived from an intra-weight image mixed with an inter-weight image.
2. The method of claim 1 wherein the intra-weight image mixed with the inter-weight image uses a minimum operation.
3. The method of claim 1 wherein the intra-weight image mixed with the inter-weight image uses a simple average operation.
4. The method of claim 1 wherein the intra-weight image mixed with the inter-weight image uses a maximum operation.
5. A robust method for image feature estimation comprising:
a. receiving at least one learning image input;
b. accumulating a weight image from the at least one learning image;
c. processing the at least one learning image using the accumulated weight image to produce a weight image output wherein the weight image is derived from mixing
i. an intra-deviation image;
ii. an intra-weight image;
iii. an inter-deviation image;
iv. an inter-weight image.
6. A robust method for image feature estimation comprising:
a. receiving at least one image input;
b. adjusting a weight image by iteration responsive to a cost function comprising
i. performing fitting using an adjusted weight image to generate a fitting
ii. determine cost function values from the fitting result;
iii. adjusting the weight image using cost function values;
iv. repeat steps i, ii, and iii until a stopping criteria is met wherein the stopping criteria is determined by the maximum allowable error,
c. estimating using the adjusted weight image to produce a fitting result.