Abstract: To improve the security of image encryption results, a high performance algorithm of image encryption is proposed by combining self-encoded theory with chaotic and super-chaotic mappings. In the mast key control condition, this algorithm firstly uses discrete improved Henon mapping through repeated iteration to generate intermediate key matrix with plaintext image size, and the binary sequence of intermediate key matrix is self-coded into Gaussian white noise sequence by using the encoder based on automate regression model, moreover it is perturbed by logistic chaotic mapping and discreted to generate random sequence with high complexity. Secondly, two different random sequences, which are respectively produced by self-encoder and two-dimensional discrete hyper chaotic mapping, are fused to generate the composite key sequence by using three invertible improved chaotic mapping of three-dimensional discrete Lorenz system. In the end, the pixel values of plaintext image are achieved the positive and negative direction diffusion encryption to generate cipher image by making good use of new two-dimensional revertible mapping with product operation over finite integer field. The experimental results show that the proposed image encryption algorithm has good cryptographic properties, and can significantly resist differential and chosen plaintext attacks.