Aiming at the problem of low embedding capacity and poor invisibility in compressed domain video hiding algorithm, a reversible steganography scheme for H.264/AVC encryption domain was proposed. Firstly, the reference frame interval parameter was determined by the embedded capacity and the carrier size, and whether the cover was encrypted was determined by the need. Then, an embedded key was generated according to the number of video frames to be embedded. Finally, the reversible information embedding on motion vector was realized by the vector histogram shifting in the compressed video. The proposed scheme overcame the distortion accumulation effect due to motion vector modification by specifying a decoding reference frame and is compatible with motion vector-based video encryption algorithms. Video decryption and information extraction depend on the decryption key and the embedded key respectively, which are separated from each other. The information can be extracted in the video ciphertext domain or the decrypted plaintext domain and has no influence on video cover recovery. As security of the information depends on the embedded key, the length of the key can be controlled as needed with the maximum length equal to the number of frames in which the information can be embedded. Experimental results show that the proposed scheme has low computational complexity and high security, and can adjust capacity and invisibility according to embedded load. Compared with BCH code reversible embedding scheme, the PSNR (Peak Signal-to-Noise Ratio) value increases by 3 to 5 dB and the average embedded capacity increases by 5 to 10 times.

According to mismatch between word semantic description of "HowNet" and subjective cognition of vocabulary, in the context of making full use of rich network knowledge, a word semantic similarity calculation method combining "HowNet" and search engine was proposed. Firstly, considering the inclusion relation between word and word sememes, the preliminary semantic similarity results were obtained by using improved concept similarity calculation method. Then the further semantic similarity results were obtained by using double correlation detection algorithm and point mutual information method based on search engines. Finally, the fitting function was designed and the weights were calculated by using batch gradient descent method, and the similarity calculation results of the first two steps were fused. The experimental results show that compared with the method simply based on "HowNet" or search engines, the Spearman coefficient and Pearson coefficient of the fusion method are both improved by 5%. Meanwhile, the match degree of the semantic description of the specific word and subjective cognition of vocabulary is improved. It is proved that it is effective to integrate network knowledge background into concept similarity calculation for computing Chinese word semantic similarity.

The parallel Particle Swarm Optimization (PSO) algorithm was improved through Graphics Processor Unit (GPU) based on Compute Unified Device Architecture (CUDA). According to the structural characteristics of the CUDA hardware system, it can be concluded that block is executed serially and the basic scheduled and executive unit of Streaming Multiprocessor (SM) is warp. GPU parallel PSO algorithm based on adaptive warp was carried out in order to make full use of thread parallelism in the block. The dimensions of particles were corresponded to the threads of particles. Each particle was corresponded to one or more warps in accordance with its self-dimension adaptively by using the warp level parallelism of GPU. One or more particles were corresponded to each block. Comparison with the existing coarse-grained parallel approach (corresponding each particle to the thread) and fine-grained parallel approach (corresponding each particle to the block) was made, and the experimental results show that the proposed parallel approach achieves CPU speed-up ratio of 40 more than two kinds of approaches mentioned above.