System for support vector machine prediction
A computer-implemented method is disclosed for image recognition and other applications. The method employs an SVM model and can reduce false negatives and increase recognition accuracies by raising the sample-to-support-vector ratio.
1. A computer system, comprising:
one or more computer processors configured to select a kernel and kernel parameters for a first Support Vector Machine (SVM) model that includes SVM support vector number of support vectors,
to test the first SVM model on a feature matrix T to produce false positive (FP) data set and false negative (FN) data set, wherein the feature matrix T includes n feature vectors of length m, wherein n and m are integer numbers,
to produce a T_best using the feature matrix T, wherein the feature matrix T_best comprises T_best sample number of sample points;
to check if a ratio (T_best sample number)/(SVM support vector number) is above a threshold for the first SVM model on the feature matrix T_best, and
if the ratio is above the threshold, to perform SVM predictions using the first SVM model on the feature matrix T_best.
2. The computer system of claim 1 , wherein the n feature vectors are based on faces or objects in images, wherein the first SVM model is used on the feature matrix T_best to classify the faces or the objects in the images.
3. The computer system of claim 2 , wherein the one or more computer processors are further configured to create an image-product design based on the faces or the objects in the images classified by the first SVM model using the feature matrix T_best.
4. The computer system of claim 1 , wherein if the ratio (T_best sample number)/(SVM support vector number) is not above the threshold for the first SVM model on the feature matrix T_best, repeating the steps of selecting, the one or more computer processors are further configured to repeat the steps of testing, automatically removing, retraining, and checking to find a second SVM model having (T_best sample number)/(SVM support vector number) ratio that exceeds the threshold.
5. The computer system of claim 4 , wherein the one or more computer processors are further configured to perform SVM predictions using the second SVM model on the feature matrix T_best.
6. The computer system of claim 4 , wherein if a SVM model having (T_best sample number)/(SVM support vector number) ratio that exceeds the threshold is not found, the one or more computer processors are further configured to select a third SVM model that yields a highest (T_best sample number)/(SVM support vector number) ratio in the step of repeating the steps of selecting, testing, automatically removing, retraining, and checking; and to perform SVM predictions using the third SVM model on the feature matrix T_best.
7. The computer system of claim 1 , wherein the threshold for the (T_best sample number)/(SVM support vector number) ratio is ten or higher.
8. The computer system of claim 1 , wherein the one or more computer processors are further configured to apply Grid Search on the feature matrix T to find optimal kernel parameters before the first SVM model on a feature matrix T is tested.
9. The computer system of claim 1 , wherein the feature vectors are based on faces or objects in images, wherein the one or more computer processors are further configured to classify the objects and the faces in the images by performing SVM predictions on the feature matrix T_best.
10. The computer system of claim 9 , wherein the one or more computer processors are further configured to automatically create designs for photo products based on the objects and faces classified by performing SVM predictions on the feature matrix T_best.
11. The computer system of claim 1 , wherein the one or more computer processors are further configured to select a feature vector v from the FP data set to remove from the feature matrix T_best to produce a feature matrix T_v,
to retrain the first SVM model on the feature matrix T_v to produce a fourth SVM model,
to calculate a benefit function to produce a benefit function value, wherein the benefit function is dependent of differences between false positives, false negatives, and numbers of support vectors generated respectively on the feature matrix T_best by the first SVM model and on T_v by the fourth SVM model,
to remove the feature vector v from the feature matrix T_best to obtain an updated feature matrix T_best if the benefit function value meets a predetermined criterion, and
to perform SVM prediction using the fourth SVM model on the updated feature matrix T_best.
12. The computer system of claim 11 , wherein the benefit function is dependent more strongly on a difference between the false negatives than on a difference between the false positives.
13. The computer system of claim 12 , wherein the difference between false negatives has a first weighting coefficient in the benefit function, wherein the difference between false positives has a second weighting coefficient in the benefit function, wherein the first weighting coefficient is larger absolute value than the second weighting coefficient.
14. The computer system of claim 11 , wherein the n feature vectors are based on faces or objects in images, wherein the fourth SVM model is used on the feature matrix T_best to classify the faces or the objects in the images.
15. The computer system of claim 14 , wherein the one or more computer processors are further configured to create an image-product design based on the faces or the objects in the images classified by the fourth SVM model using the feature matrix T_best.
16. The computer system of claim 11 , wherein the one or more computer processors are further configured to repeat the steps of selecting a feature vector v, training the first SVM model, performing SVM predictions, calculating a benefit function, and removing the feature vector v from the feature matrix T_best by selecting and removing a different feature vector corresponding to the FP data set from feature matrix T_best,
wherein SVM predictions are performed using the feature matrix T_best that gives a highest best function value.
17. The computer system of claim 16 , wherein all the feature vectors corresponding to the FP data set are evaluated by calculating a corresponding benefic function and determined to be removed from the feature matrix T_best or not depending on a value of the corresponding benefic function.
18. The computer system of claim 11 , wherein the feature vector v is not removed from the feature matrix T_best if the value of the benefit function does not meet the predetermined criterion.
19. The computer system of claim 11 , wherein the feature vectors are based on faces or objects in images, wherein the one or more computer processors are further configured to classify the objects and the faces in the images by performing SVM predictions on the feature matrix T_best.