Abstract
A new retrieval technique of salient image features is proposed in this research and this technique firstly detects the feature points by using Harris corner detector to find out the feature region of image. And then compute the mean of color, variance, the number of occurrence of color and color aggregation and any eigenvector in accordance with the chrominance and lightness of feature region to resist the geometric attacks existing in image. Afterwards, compute the similarity by matching the feature region and resend the most similar image to user. For the experimental result, different types of animal databanks and Caltech 101 human face databanks are tested and the precision and recall of image retrieval are both good. Thus, this technique can retrieve similar image effectively.