This study proposes the Jamming Image Recognition (JIR) Problem and a perspective solution with hybrid features. JIR aims to recognize the prototypical image in each class and differentiate two prototypical images for two arbitrary classes. Each class was generated by a prototypical image with some jamming skills—rotation, center cropping, edge cropping, translational treatment, gamma color correction and etc. It is a prototypical recognition problem (matching with prototypical images) and generalized from Face Recognition Problem (matching with prototypical faces). As a perspective solution, we optimize the oriented Features from Accelerated Segment Test and rotated Binary Robust Independent Elementary Features (ORB) matching algorithm by the discrete cosine transform (DCT) model and restoration algorithms. The ORB-matched feature matrices are utilized to avoid incorrect matching and then, subjected to DCT transformation. Subsequently, the upper left corner submatrices are compared and reassigned to extract hybrid features for final recognition. This novel method was validated on 2100 classes of macroscopic remote-sensing images and 1926 classes of microscopic fundus images. With a high similarity threshold Thr = 0.9, the recognition rates on macroscopic images (accuracy = 97.16%) and microscopic ones (accuracy = 85.83%) are much higher than perceptual Hash (pHash) (62.73% ~ 73.69%) and ORB (0% ~ 24.21%), demonstrated that our method is effective. In subsequent studies, a next research priority is to reduce the time cost (1.8 s per image) in JIR with hybrid features.