Research Article
Multilabel Image Annotation Based on Double-Layer PLSA Model
Algorithm 1
The image concept detection algorithm.
(1) Input: a new untagged image | (2) Get the BoW representation of , a dimensions vector: | | (3) According to the training parameter , applying the folding-in algorithm to get | the visual latent topic distribution: | | (4) The same as the step (3), according to , using the folding-in algorithm to get | the top-layer latent topic distribution of image : | | (5) Combine the training parameter and the top-layer latent topic distribution | gained from the step (4) to get the label latent topic distribution of image : | | (6) According the training parameter and the distribution of , the probability of | each image semantic label appearing on the image can be calculated: | | then choose ( can be selected as need, in this paper ) labels with the largest | probabilities to construct the label set of image | (7) Output: the labels of image |
|