Perceptual image quality is one of the most important performance indicators for digital TVs. In order to provide pleasing impression on viewers, preferred color correction that modifies memory colors is often applied. The first step of the preferred color correction is to extract pixels of memory colors. This paper presents a new method to determine the optimized boundary of geometrical figures for memory color extraction. Genetic algorithm is utilized for optimization. Based on relationship between lightness and hue and/or chroma, range of lightness is subdivided into three groups. Boundary of geometrical figure is determined for each group. Performance of the proposed method is evaluated by calculating the extraction accuracies. Experimental results indicate that proposed method outperforms previous methods proposed for the memory color extraction.