SKCS - SEPARABLE KERNEL FAMILY WITH COMPACT SUPPORT (TA-L1)

Author(s) : Ezzedine Ben Braiek (ESSTT, Tunisia) Mohamed Cheriet (ETS, Canada)

Abstract : In this paper, we focus in improving the efficiency of our newly introduced kernel family with compact support (KCS, [1]), by proposing a separable version of KCS, namely SKCS, which itself has a compact support with nice features. The Laplacian of this new kernel family is written as a sum of two one-dimensional filters, leading to perform the operator in a separate way, following x and y directions. This contributes to better extraction of pertinent data from gray level document images with noisy backgrounds, which remains a challenging problem in character recognition applications. The new method achieved good performance while reducing drastically time processing. Comparative results show the relevance of the new kernel family and its superior performance.

Menu