Based on the analysis of the advantages and disadvantages of the existing algorithms, various types of images are considered as far as possible. Based on the principle prototype of light emitting diode-liquid crystal display direct-down image and video display, a reliable and effective data measurement method based on regional backlight brightness extraction is proposed, and an efficient and practical backlight brightness extraction method based on deep learning is proposed. The method is based on the neural network with a multi-layer down sampling structure to extract the image features and obtain the optimal backlight. The experimental results show that the method can improve the image display quality and expand the dynamic range of images. The experimental results of network structures with or without bypass verify the superiority and effectiveness of the proposed method.