Table 2.

Parameters of the classification algorithms used for extracting visual cues.

Type of visual cues	Algorithm	Parameters
Color-oriented	Color histogram intersection	Type of color space: RGB, HSV Classifier: KNN Distance: correlation
Texture-oriented	BVW approach	Classifier: SVM with Gaussian kernel Interest points detectors: SIFT Feature representation: SURF Codebook generation: KNN
Instrument categorization	Viola-Jones approach	Features: Haar-like rectangular Negative images: 2000 Positive images: 500
Detection of other instruments	BVW approach	Classifier: SVM with Gaussian kernel Interest points detectors: SURF Feature representation: SURF Codebook generation: KNN
Alternative method	Global features classification	Spatial features: RGB, HSV spaces, Haralick descriptors, DCT, spatial moments Wrapper method: RFE-SVM Filter method: MI Classifier: SVM with Gaussian kernel