In order to solve the problems that SIFT(Scale Invariant Feature Transform, SIFT) descriptor may result in a lot mismatches when an image has many similar structures and its high dimensions will consume much time in image matching. This paper presents a compressive sensed SIFT descriptor which is mixed with relative geometry location. At first，this method centers on feature point，and transforms the information of relative geometry location related to around key points into a RGL(Relative Geometrical Location, RGL) descriptor, which is invariant to scale and rotation. Secondly, CS-SIFT(Compressive Sense SIFT, CS-SIFT）descriptor is formed by reducing dimensions of SIFT descriptor using the theory of compressive sense. At last, two descriptors form a RGL-CS-SIFT descriptor(descriptor mixed with RGL and CS-SIFT, RGL-CS-SIFT). The results indicate that the RGL-CS-SIFT increases the matching speed and improves the ratio of image matching significantly, compared with SIFT and PCA-SIFT(Principal Component Analysis SIFT, PCA-SIFT) descriptors.