In this paper we address the problem of learning to recognize objects by manipulation in a developmental robotics scenario. In a life-long learning perspective, a humanoid robot should be capable of improving its knowledge of objects with active perception. Our approach stems from the cognitive devel- opment of infants, exploiting active curiosity-driven manipulation to improve perceptual learning of objects. These functionalities are implemented as perception, control and active exploration modules as part of the Cognitive Architecture of the MACSi project. In this paper we integrate these functionalities into an ac- tive perception system which learns to recognise objects through manipulation. Our work in this paper integrates a bottom-up vision system, a control system of a complex robot system and a top-down interactive exploration method, which actively chooses an exploration method to collect data and whether interacting with humans is profitable or not. Experimental results show that the humanoid robot iCub can learn to recognize 3D objects by manipulation and in interaction with teachers by choosing the adequate exploration strategy to enhance competence progress and by focusing its efforts on the most complex tasks. Thus the learner can learn interactively with humans by actively self- regulating its requests for help.