Creating cartoon is a challenging and time-consuming task for artists. However, automated technology that converts real photos into high-quality cartoon-style images is significantly valued. Therefore, based on a generative adversarial network, this study proposes a lightweight image cartoon stylization method. By observing the cartoon drawing behavior, the cartoon image style is decoupled into three representations, including smooth surface, sparse color block, and high frequency texture. A generative adversarial network framework is used to learn the extracted representation and the style of cartoon images. Furthermore, deep detachable convolution and reverse residual blocks are used in generative networks to reduce number of network parameters and computational costs. Qualitative comparison and quantitative analysis are conducted in this study to evaluate the proposed method's effectiveness. The results show that the proposed method can quickly convert real-world photos into high-quality cartoon images and is superior to the existing methods.