Similarity Analysis of Digital Image with Nonparametric Tests on Time Series

Similarity search is concerned with efficiently locating subsequences or whole sequences in large archives of sequences. Since the multimedia data can be easily represented as a time series the concept of time series similarity search can be easily extended to compute the similarity between two digital images. Several distance measures such as Euclidean distance, Earth Mover’s Distance (EMD), etc have been  used in finding the similarity between two given time series. In the proposed work, time series similarity analysis that uses nonparametric test statistics is adopted to find similarity between the given images. Initially the given images are transformed into time series and its dimensionality is reduced. The resultant time series is represented as clusters by the use of k-means clustering and the similarity distance between two images is found using NonParametric Tests (NPT). Also, a Composite Similarity Measure (CSM) that comprises of EMD and nonparametric tests is proposed. The experimental results show that the proposed measure is well suited for measuring the subjective similarity between two images.