3 ms·

I think that if I were to do this, I would use facial landmark recognition (using something like a Haar classifier). Haar-like features have been used to aid in

by dvt 13y ago

I think that if I were to do this, I would use facial landmark recognition (using something like a Haar classifier). Haar-like features have been used to aid in (human) facial recognition since 2001 to great success[0]. And recently, people have been thinking about using similar methods for animal tracking[1]. If one could locate the face in the test set, she could also presumably find some landmarks of interest: eyes, nose, mouth, etc. Considering that dogs typically have longer snouts, cats have pointier ears, etc, this data could be used to differentiate between a dog and a cat. There would be difficulty dealing with awkward angles and bad lighting though. [0] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.6.3549 http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.6.35... [1] http://www.eng.auburn.edu/~troppel/internal/sparc/TourBot/TourBot%20References/Haar/2000186.pdf http://www.eng.auburn.edu/~troppel/internal/sparc/TourBot/To...

apu 13y ago
Haar wavelets are most useful for detecting faces (drawing a rectangle around the entire face). They are not very good for locating landmarks on the face. Also, they tend to be much more sensitive to the orientation of the face than other features, so modern face detectors are often composed of multiple independent detectors, each specialized for different pose angles. Standard computer vision features like HOG (histograms of oriented gradients) or SIFT will probably do much better, or the deep learning features others have mentioned. Your larger point of adapting a face detectors for animal use is well taken, though probably overkill for simply saying "dog" or "cat". You need that level of detail to identify which breed (e.g., this is the approach that dogsnap takes), but not for the base distinction. The other way to go would be to train a deformable parts model (DPM) detector [1] for dogs and cats. DPMs are the current state of the art in detecting objects, e.g. as measured on the pascal VOC benchmark [2]. [1] http://www.cs.berkeley.edu/~rbg/latent/ http://www.cs.berkeley.edu/~rbg/latent/ [2] http://pascallin.ecs.soton.ac.uk/challenges/VOC http://pascallin.ecs.soton.ac.uk/challenges/VOC

by dvt 13y ago

apu 13y ago

Haar wavelets are most useful for detecting faces (drawing a rectangle around the entire face). They are not very good for locating landmarks on the face. Also, they tend to be much more sensitive to the orientation of the face than other features, so modern face detectors are often composed of multiple independent detectors, each specialized for different pose angles. Standard computer vision features like HOG (histograms of oriented gradients) or SIFT will probably do much better, or the deep learning features others have mentioned. Your larger point of adapting a face detectors for animal use is well taken, though probably overkill for simply saying "dog" or "cat". You need that level of detail to identify which breed (e.g., this is the approach that dogsnap takes), but not for the base distinction. The other way to go would be to train a deformable parts model (DPM) detector [1] for dogs and cats. DPMs are the current state of the art in detecting objects, e.g. as measured on the pascal VOC benchmark [2]. [1] http://www.cs.berkeley.edu/~rbg/latent/ http://www.cs.berkeley.edu/~rbg/latent/ [2] http://pascallin.ecs.soton.ac.uk/challenges/VOC http://pascallin.ecs.soton.ac.uk/challenges/VOC