In classical image-based visual servo (IBVS) control systems, image Jacobian matrix estimation depends on target-depth information. To address this concern, the proposed study describes use of a novel control-system approach, wherein multivariate adaptive regression splines (MARS) are first used to select important features of a dataset to reduce training-sample size. Subsequently, a hybrid algorithm introduces random reduced kernel regularization (RKR) parameters based on the use of an incremental extreme learning machine (IELM). The said RKR-IELM algorithm solves the singularity problem that occurs when the number of training samples is less compared to the number of hidden-layer neurons within the kernel extreme learning machine algorithm. Furthermore, RKR-IELM reduces the risk of overfitting. Finally, a hybrid algorithm that combines MARS with RKR-IELM has been proposed for use in an IBVS control system. The said hybrid MARS-RKR-IELM algorithm can solve the classic problem of an IBVS system associated with estimation of the image Jacobian matrix and poor real-time performance. Excellent performance of the proposed system has been verified via comparisons performed against other related systems.