A Three Dimension Super-Resolution Algorithm through Neighbor Embedding Based on Weighted Coefficient Values for Internet of Things

In the applications of the internet of things (IOT), the nodes in the network usually has power limits and storage space limits. Thus, the images over the IOT are usually low-resolution images. However, the customer usually want high-resolution images. To satisfy these needs, the solution of super-resolution methods comes. In this paper, a new algorithm for super-resolution is proposed, where the three dimensional discrete cosine transform (3D-DCT) is employed to extract the features of the image blocks. Then, the low frequency coefficients in the 3D-DCT are set with large weight values, and the high frequency coefficients are set with relatively small weight values. The 3D-DCT coefficients are then multiplied with the corresponding weights, and a reverse 3D-DCT is carried out. A bi-cubic transform is employed to get a high resolution block. The difference between that block and the actual high resolution block is saved in the training set with the 3D-DCT blocks. A new difference measure is proposed in the online phase to obtain the similarity between a candidate block and the blocks in the training set. After this, several training blocks are selected and the neighbor embedding method is employed to reconstruct the high resolution blocks. To reduce the computational complexity of the algorithm, the revised K-means algorithm is employed. The experimental results show that the proposed algorithm performs much better than the traditional neighbor embedding algorithm and the bi-cubic interpolation algorithm.