CMOS image sensors have been extensively used in applications that carry out image acquisition and conversion to the digital domain. The sensors have been extensively used in real-time applications that require them to be efficient in saving power. Off-chip processing of images either requires it to process over the cloud or use a general computational resource such as a CPU or a GPU. For making sensors more power-efficient a CNN model is developed that can be engraved on-chip with the sensor for image reconstruction after reading only a fraction (15 to 10%) of the pixels. The design is to be implemented and configured for the proposed CNN architecture.