learning phase of Viola Jones / AdaBoost

Question

I have a Problem understanding the trainingsphase of the Viola Jones algorithm.

I give the algorithm in pseudo code, as far as I understand it:

# learning phase of Viola Jones
foreach feature # these are the pattern, see figure 1, page 139
    # these features are moved over the entire 24x24 sample pictures
    foreach (x,y) so that the feature still matches the 24x24 sample picture
        # the features are scaled over the window from [(x,y) - (24,24)]
        foreach scaling of the feature
            # calc the best threshold for a single, scaled feature
            # for this, the feature is put over each sample image (all 24x24 in the paper)
            foreach positive_image
                thresh_pos[this positive image] := HaarFeatureCalc(position of the window, scaling, feature)
            foreach negative_image
                thresh_neg[this negative image] := HaarFeatureCalc(position of the window, scaling, feature)
            #### what's next?
            #### how do I use the thresholds (pos / neg)?

This is, btw the frame as in this SO Question: Viola-Jones' face detection claims 180k features

This algorithm calls the HaarFeatureCalc-function, which I think I understood:

function: HaarFeatureCalc
    threshold := (sum of the pixel in the sample picture that are white in the feature pattern) -
        (sum of the pixel in the sample picture that are grey in the feature pattern)
    # this is calculated with the integral image, described in 2.1 of the paper
    return the threshold

any mistakes till now?

The learning phase of Viola Jones, basically detects which of the features/detectors are the most deciding. I don't understand how the AdaBoost works, that is described in the paper.

Question: how would the AdaBoost from the paper look like in pseudo code?

ask ml related questions in metaoptimize.. For this question is suited more there :) — frazman, Apr 15 '12 at 21:42
I did, http://metaoptimize.com/qa/questions/9931/learning-phase-of-viola-jones-adaboost — Jörg Beyer, Apr 16 '12 at 19:55

Jai · Answer 1 · 2018-01-08T20:55:43.287

Viola Jones :

First you will hard code 180k classifiers(feartures)
And initially all the training data will have same weights
Now each classifier(feature) will be applied to every training data you have
Each feature will classify your training data as either a face or not a face
based on this classifications for every feature you will calculate the error
After calculating the error of each feature, you pick the feature with error that is most far away from 50% by which i mean if you have one feature with 90% error and other feature with 20% error then you will select the feature with 90% error and add it to the strong classififer
You will update weights of each training data
Now you will repeat the process untill you get a nice accuracy for you validation data with the strong classifier that you have built
AdaBoost is a technique of making a strong classifiers out of weak classifier

learning phase of Viola Jones / AdaBoost

1 Answers1