Image interpolation is an effective method to magnify image with low resolution to adapt to the target screen. Better the preservation of image edges brings much better magnifying result. A new adaptive image interpolation approach based on geometric classification of interpolate units

which is formed by edges of image with low resolution

is proposed. Firstly

we construct rectangle interpolate units and rhombic interpolate units by the relationship of pixels in image with high resolution. All pixels of the superresolution image lie in the two kinds of interpolate units. Then we analysis possible edges of original images especially along the oblique line and treat edges as the secants of interpolate units to classify every interpolate unit. There are total 16 kinds of interpolate units. At last

we calculate unknown pixels based on the classified interpolate units. Experiments show that our new adaptive image interpolation scheme based on geometric classification can preserve edges of interpolated images better than many present methods.