In the field of image distortion, the difficulty of image annotation is mainly reflected in three points:(1)The appraiser's evaluation annotation is inconsistent; (2)The boundary definition of the image distortion type is fuzzy;(3)The environment atmosphere of the tagging work is complex. Three kinds of difficulties are often caused by the ambiguity in the distortion of image annotation results. As an effective solution to ambiguity and uncertainty. In this paper, a semi-automatic method based on neighborhood rough sets is proposed for distorted images. The aim of this paper is to improve the accuracy of annotation by constructing a global rough set model. Specifically, under the constraint of defined annotation rules. The sample is annotated by manual annotation, and the approximate neighborhood of the sample is constructed. Then according to the approximate neighborhood coordinates, calculate the coordinates in the neighborhood of upper approximation and lower approximation. Finally, construct the semantic association between the annotation words and the images, so as to classify the images. The experimental results show that the method has achieved effective results in image distortion classification.
[1]
Shih-Fu Chang,et al.
To search or to label?: predicting the performance of search-based automatic image classifiers
,
2006,
MIR '06.
[2]
Xinlei Chen,et al.
Microsoft COCO Captions: Data Collection and Evaluation Server
,
2015,
ArXiv.
[3]
Marcel Worring,et al.
Unsupervised multi-feature tag relevance learning for social image retrieval
,
2010,
CIVR '10.
[4]
Saul B. Gelfand,et al.
Classification trees with neural network feature extraction
,
1992,
IEEE Trans. Neural Networks.
[5]
Grigorios Tsoumakas,et al.
Multi-Label Classification: An Overview
,
2007,
Int. J. Data Warehous. Min..
[6]
Rich Caruana,et al.
Classification with partial labels
,
2008,
KDD.