Secure access control applications like border control rely on the face based verification system by considering its reliability, usability and accuracy in-person verification. However, face recognition systems are vulnerable to morphed face attacks, in which, the morphing process combines two different facial images into a single facial image. The features extracted from the morphed face image will match to those extracted from probe images of both faces. Thus, it is essential to reliably detect the morphed face image attacks on the face recognition systems. In this work, we propose a novel approach to detect morphed face images using residual color noise. The proposed method is designed to capture the noise patterns that are a result of the morphing process. Thus, the proposed method performs first denoising using Deep Convolutional Neural Network (CNN) independently on the Hue Saturation Value (HSV) color space, and then computes the residual noise. The extracted residual noise is further processed using Pyramid Local Binary Patterns (P-LBP), which is further classified using the Spectral Regression Kernel Discriminant Analysis (SRKDA). Extensive experiments are carried out on three different morphed face image datasets. The Morphed Attack Detection (MAD) performance of the proposed method is benchmarked with 13 different state-of-the-art techniques using the ISO IEC 30107-3 evaluation metrics. Based on the obtained quantitative results, the proposed method has indicated the best performance.