Image processing is a type of memory-access-intensive application that requires to access memory for a longer duration thereby lowering the overall throughput of the system by limiting the speed of memory access and bandwidth. This profound memory wall problem can be overcome by using memristor-based Computation-In-Memory (Memory-driven) architecture that can be simultaneously utilized as memory and processing element. To further enhance the computational speed and to perform energy/power-efficient operations, approximate computing is employed by allowing leverages in output for certain combination of inputs. In image processing applications, the computing blocks consist of multipliers and delays whose working is determined by suitable combination of adders. In this work, memristor-based approximate full adder is designed by random incorporation of errors in SUM and CARRY, respectively, and the resulting truth table is verified. The capability of proposed approximate adder is validated by designing (in Cadence Virtuoso) a 4-bit ripple carry adder (RCA) to perform bit-wise pixel addition of two grayscale images of same size and compared with an image obtained by exact addition method. © 2018, Springer Nature Singapore Pte Ltd.