Kernel methods and scale invariance using the triangular kernel

Hichem Sahbi (1) , François Fleuret (1)

1 IMEDIA - Image and multimedia indexing, browsing and retrieval

We focus in this paper on the scale invariance of kernel methods using a particular function referred to as the triangular kernel. The study in (Sahbi and Fleuret, 2002) reported scale invariance for support vector machines (SVM) and the current work is an extension for support vector regression (SVR) and kernel principal component analysis (KPCA). First, we review these kernel methods and we illustrate analytically the scale invariance of the training processes. Experiments are conducted in several cases showing the scale invariance and the good performance in real pattern recognition problems including shape description, face detection and recognition.

Mots clés: STATISTICAL LEARNING; KERNEL METHODS; SCALE-INVARIANCE; TRIANGULAR KERNEL; SHAPE DESCRIPTION; FACE DETECTION AND RECOGNITION

Domaines: Autre [cs.OH]

Hichem Sahbi, François Fleuret. Kernel methods and scale invariance using the triangular kernel. [Research Report] RR-5143, INRIA. 2004. ⟨inria-00071440⟩