Nowadays, research essentially tends to overcome the power consumption issue in the digital and analog domains. In this paper, we took a closer look on CMOS image sensors. The sensor consumption mainly depends on the image resolution. In other words, the sensor data flow is always constant regardless the activity or the scene in question. This leads to a great amount of useless data to process, store and later on display. Therefore we rethought the image sensor functioning as well as its reading method in order to reduce the sensor data flow and thus its power consumption. A novel asynchronous image sensor has been proposed using a new sampling scheme, a specific processing and an asynchronous reading technique. On one hand, the idea mainly consists of limiting the processing by only extracting the relevant data. On the other hand, the sampling method eliminates the spatial redundancies in the captured scene. Finally, by implementing the proposed reading system, we were able to drastically reduce the sensor data flow and, consequently, to design a power-efficient asynchronous image sensor.