Localized Multiple Kernel Learning (LMKL) Algorithm is a kind of multiple kernel support vector machine algorithm with varying coefficient and uses gating function to locally select suitable compound kernel function, however, its gating function has serious parameter redundancy, therefore, this paper proposes improved localized multiple kernel learning (ILMKL) algorithm, adds regularization term to objective function to discriminate l1 norm form of the gating function in LMKL, uses lp norm form of gating function to enhance “complementary” role between kernel functions, uses this algorithm to conduct experiments on simulation data set and UCI data set and draws the conclusion that this algorithm gets higher classification capacity. 