The normal field of the scene can be obtained by various methods such as shadow shape, deflection measurement and photometric stereo. The height field function obtained from the normal field plays an important role in 3D reconstruction methods such as shadow shape and photometric stereo, so the depth information of the scene reconstructed from the normal field is worth studying. A method to recover height field function from discrete normal vector is proposed in this paper. Firstly, based on the principle of discrete geometry, the row height value of the scene is estimated line by line. Then, a small number of column heights are calculated. Finally, according to the column height value, the average height of each row is adjusted row by row to obtain the height distribution of the whole scene. The least square method is used in the process of estimating the row height value and column height value. This method reduces the amount of optimization data. It does not need large matrix operations and operation memory, and has a good effect on the surface of the object, which is continuous and smooth. Experimental results show that the depth recovery of the scene with few continuous surface mutations is better, but for the scene with many mutations, the recovered height will have horizontal stripes.