In order to recover depth information from monocular infrared image, a depth estimation algorithm based on novel deep convolutional neural networks (DCNN) is proposed. The texture energy and texture gradient of infrared images are extracted by using Laws′ masks and the gradient detector at different scales. These two types of texture information are considered as the first kind of features. The selected gray values and their statistical histogram in specific areas are considered as another two kinds of features. The DCNN are trained on these three kinds of features with the corresponding depth labels respectively. The trained DCNN are then utilized to estimate the depths of testing monocular infrared images respectively. Experimental results show that compared with other methods, the DCNN trained by texture information can estimate the depth much better than those of the existing methods, especially in the depth changes of local scenes.