This paper proposes an optimized multicarrier (MC) spread spectrum (SS) image watermarking scheme using hybridization of genetic algorithms (GA) and neural networks (NN). Data embedding is done in the mutually independent host components using the distinct code patterns that are assigned to the different watermark bits. GA determines the gradient thresholds for the pixel intensities to partition the host image into the edge, the smooth, and the texture regions as well as determines the watermark embedding strengths. The goal is to optimize the imperceptibility and the data hiding capacity. A minimum mean square error combining (MMSEC) decoder is used and the weight factors are calculated using NN through training/learning. Stable decision variables thus obtained for the watermark bit detection are partitioned into the multiple groups to improve decoder performance by canceling out the multiple bit interfering effect. Simulation results show the relative performance gain achieved in this method compared to the existing other works including the biologically inspired approaches.