Image Matching Using Pseudo Time Series Representation

Boundary and edge are highly beneficial in improving a wide variety of vision tasks such as semantic segmentation and object recognition. While classical edge detection is a challenging problem in itself, the semantic edge detection is more challenging problem. The classical edge detection task has been shown beneficial for solving many computer vision tasks such as 3d reconstruction [1], 3d shape recovery [2], medical image processing [3], as well as semantic segmentation [4,5].This paper proposes two pre-processing techniques for image/object retrieval. A conventional pre-processing technique is used to define the object as a pseudo time series in one dimension. The suggested first technique creates modified versions of the SAX representation: it employs an approach known as Extended SAX (ESAX) to achieve efficient and accurate discovery of key patterns, which is required for finding the most plausible related items. The similarity between two images/objects is then defined as the overall similarity between two families of symbolic words. A distance measure is used to decide the most plausible matching between strings of symbolic words. We empirically compare the Extended SAX with the original SAX approach and demonstrate its improvement in retrieving the most plausible similar objects at the higher cardinality.

The second technique denotes each object/shape by a specific set (subset) of its boundary points. Each point is a center of a small region around it that is located as an image patch reflecting the low-level features of that patch. The image/object is associated with a family of image/object features corresponding to its patches. The similarity between two images/object is then defined as the overall similarity between two families of image/shape patches. GA is applied to decide the most plausible matching. The experimental results have shown that our approachs is effective in retrieving similar images.