A new image size reduction model for an efficient visual sensor network

Abstract Image size reduction for energy-efficient transmission without losing quality is critical in Visual Sensor Networks (VSNs). The proposed method finds overlapping regions using camera locations, which eliminate unfocussed regions from the input images. The sharpness for the overlapped regions is estimated to find the Dominant Overlapping Region (DOR). The proposed model partitions further the DOR into sub-DORs according to capacity of the cameras. To reduce noise effects from the sub-DOR, we propose to perform a Median operation, which results in a Compressed Significant Region (CSR). For non-DOR, we obtain Sobel edges, which reduces the size of the images down to ambinary form. The CSR and Sobel edges of the non-DORs are sent by a VSN. Experimental results and a comparative study with the state-of-the-art methods shows that the proposed model outperforms the existing methods in terms of quality, energy consumption and network lifetime.