Noise in the image are random variations in intensity due to intrinsic or extrinsic sources. This paper proposes a high throughput Fixed Point Discrete Kalman filter (DKF) architecture for denoising images with additive white Gaussian noise (AWGN) at real-time. A linearized state model based on neighbor pixel similarity is used for improving the PSNR of the noisy image. A 5-stage two parallel bi-functional trapezoidal systolic array architecture based on modified Faddeev algorithm (MFA) is chosen as basic functional block of DKF. Two parallel sets of MFA unit are utilized to increase the throughput by 1.6×, reduce the latency by 1.8× while trading off an increased utilization. In addition, the boundary cell of MFA is modified by replacing the divider unit with a highly pipelined two stage Lookup table based Newton Raphson divider. The architecture implemented on Xilinx Virtex-6 FPGA can denoise 512 × 512 images in real-time (≈ 33 fps) achieving highest throughput among the state of art architectures. A quantitative and qualitative evaluation of denoising on synthetic and real world images shows the applicability of the proposed architecture.