The work aims to propose a compact image hash algorithm based on fusion robust features and multidimensional scaling transform, in order to enhance the robustness and recognition ability of hash sequences for arbitrary rotation angles. Firstly, bilinear interpolation was used to fix the hash sequence length of the image for obtaining images of regular size. The Gauss low-pass filtering was used to filter regular images to eliminate the influence of noise pollution and interpolation error on the hash generation. Then, the filtered image was transformed into YCbCr color space for extracting the brightness Y component, so as to enhance the hash robustness with brightness adjustment. Then, the log-polar transform LPT was used to process the brightness Y component for outputting the secondary image. The SVD mechanism was introduced to decompose the secondary image for obtaining the robust features of its anti-rotation. At the same time, the local salient features of the Y component were obtained based on Fourier transform and residual mechanism. Subsequently, the two features were combined to form a fusion robust feature, which was regarded as an intermediate hash sequence. The multidimensional scaling was introduced to compress the intermediate hash sequence for obtaining compact hash. The compact hash sequence was encrypted base on Logistic map to form the target hash. By calculating the Hamming distance of hash sequence between the real image and the image to be authenticated, the image recognition was determined according to the preset threshold. Experimental results showed that, the proposed algorithm had stronger robustness and more compact hash length, as well as more ideal ROC curve, compared with the existing hash scheme, and its correct recognition rate of stability was over 96% under a variety of attacks. Featured by good robustness and sensitivity, the proposed hash scheme has better application value in packaging icon retrieval, information watermarking and other industries.