A novel fog-degraded image restoration model is proposed. This model is comprised of two parts: the defogging and the enhancement. The defogging work uses the three-dimensional color model to correct the weight of the multiscale retinex algebraic model and determine the golden defogging surround scale by the golden section. The enhancement part firstly marks the region of interesting (ROI) by double template, four-dimensional projection and morphological method. Then, the illumination is equalized in the frequency domain of the image, and the color fidelity is improved by power transformation. Subsequently, ROI is enhanced by adaptive piecewise linear method. Eventually, the enhanced ROI and the non-ROI are combined together by linear weight to generate the final result. The sufficient synthetic and real experiments prove that the proposed method has the satisfactory effects on the image detail recovery, the color fidelity and the balanced image vision in different image depth, thus meeting the demands of image monitoring and recognition.