An adaptive readout scheme for low-power CMOS image sensors is proposed in this paper. In the first step, the entire pixel array is partitioned into square blocks. Subsequently, the intensity range of each block is divided into a certain number of bins where the bin thresholds are adapted to the intensity range of the block. For this purpose, the maximum and minimum intensity values in each block are sensed by the winner take all circuit which is a common module used in hardware implementations of neural networks. The proposed scheme is implemented and simulated as a proof of concept and verified on standard images. The observed peak signal-to-noise ratio for the cameraman and thermal image is 40.37 dB and 41.16 dB, for 8×8 block size, respectively with 70.1% saving in the total power compared to the conventional readout. An additional power saving of 16% in the readout is observed by reducing the total number of threshold levels.