Like humans and primates, artificial creatures like robots are limited in terms of allocation of their resources to huge sensory and perceptual information. Serial processing mechanisms are believed to have the major role on such limitation. Thus attention control is regarded as the same solution as humans in this regard but of course with different attention control mechanisms than those of parallel brain. In this paper, an algorithm is proposed for offline learning of top-down object based visual attention control by biasing the basic saliency based model of visual attention. Each feature channel and resolution of the basic saliency map is associated with a weight and a processing cost. Then a global optimization algorithm is used to find a set of parameters for detecting specific objects. Proposed method is evaluated over synthetic search arrays in pop-out and conjunction search tasks and also for traffic sign recognition on cluttered scenes.