The figures below show learned image templates as active basis template or adaboost classifiers (with 400 negative examples). In each figure, the top-left template illustrates the selected Gabor elements, where each selected Gabor element is represented by a bar of the same size. Following the template we show in pairs the original example image together the deformed template for that particular image. These example images and deformed templates are ranked by their likelihood scores in descending order. The weak classifiers of the adaboost template are created by thresholding Gabor responses. A red bar means the weak classifier is true when the response is larger than a threshold. Otherwise it is denoted by a blue bar.
The following figure illustrates the active basis template learned from the cow category. The template is made up of 75 Gabor elements of scale 0.7. The bounding window of the template is 100 by 140 pixels.
The following figure illustrates the adaboost strong classifier learned from the cow category. The strong classifier is made up of 75 weak classifiers on locally maximized Gabor responses. The scale of Gabor elements is also 0.7. The bounding window of the strong classifier is 100 by 140 pixels.
The following figure illustrates the active basis template learned from the deer category. The template is made up of 60 Gabor elements of scale 0.7. The bounding window of the template is 120 by 120 pixels.
The following figure illustrates the adaboost strong classifier learned from the deer category. The strong classifier is made up of 60 weak classifiers on locally maximized Gabor responses. The scale of Gabor elements is also 0.7. The bounding window of the strong classifier is 120 by 120 pixels.
The following figure illustrates the active basis template learned from the cat head category. The template is made up of 75 Gabor elements of scale 0.7. The bounding window of the template is 120 by 120 pixels.
The following figure illustrates the adaboost strong classifier learned from the cat head category. The strong classifier is made up of 75 weak classifiers on locally maximized Gabor responses. The scale of Gabor elements is also 0.7. The bounding window of the strong classifier is 120 by 120 pixels.
The following figure illustrates the active basis template learned from the wolf head category. The template is made up of 60 Gabor elements of scale 0.7. The bounding window of the template is 120 by 120 pixels.
The following figure illustrates the adaboost strong classifier learned from the wolf head category. The strong classifier is made up of 60 weak classifiers on locally maximized Gabor responses. The scale of Gabor elements is also 0.7. The bounding window of the strong classifier is 120 by 120 pixels.
The following figure illustrates the active basis template learned from the horse category. The template is made up of 60 Gabor elements of scale 0.7. The bounding window of the template is 100 by 125 pixels.
The following figure illustrates the adaboost strong classifier learned from the horse category. The strong classifier is made up of 60 weak classifiers on locally maximized Gabor responses. The scale of Gabor elements is also 0.7. The bounding window of the strong classifier is 100 by 125 pixels.
The following figure illustrates the active basis template learned from the butterfly category. The template is made up of 60 Gabor elements of scale 0.7. The bounding window of the template is 100 by 150 pixels.
The following figure illustrates the adaboost strong classifier learned from the butterfly category. The strong classifier is made up of 60 weak classifiers on locally maximized Gabor responses. The scale of Gabor elements is also 0.7. The bounding window of the strong classifier is 100 by 150 pixels.