A generalized noise reconstruction approach is proposed for improved direction-of-arrival (DOA) estimation. First, coarse estimation of the noise component in the received data is obtained based on cumulative sum of approximate first-order finite difference series, where the generalized pattern search (GPS) algorithm is used to find the best initial value. The objective function is constructed with the total noise power estimated by the least squares (LS) method. After that, a new cumulative sum constructed with the solution from the previous step is used to accurately reconstruct the noise component in the data. Finally, robust DOA estimation is achieved through the denoised data. In simulations, different cases of noise, such as non-uniform noise, colored noise, impulsive noise, and K-distribution sea clutter, are considered. Real-world frequency-modulated continuous wave (FMCW) multiple-input multiple-output (MIMO) radar data is also used to demonstrate the effectiveness of the proposed approach.