Image registration technique has been widely studied and used in many fields, such as computer vision, remote sensing, medical diagnosis and treatment, environmental monitoring. Presently, registration happens after the color image is transformed into gray one in most algorithms, which causes color information loss and may lead to wrong registration. For this reason, a color image registration algorithm based on color scale invariant feature transform (CSIFT) is proposed. Firstly, color invariant value is calculated at each location in color images. Secondly feature points are extracted and neighbor information around these points is described using color invariant value as input information. Thirdly the points between two images are matched using the nearest neighbor method. Finally transformation parameters between images and the registered image can be determined using matched features. Experimental results indicate that high accuracy and small errors can be achieved when parameter values are transformed in color images, and mapping relationship can also be acquired correctly in the condition of unknown transformed relationship. This method is fast in operation in most conditions.