Region of Interest Extraction Method Using Wavelets

In this work, the Full-Wavelet Region of Interest Extraction Method (FWREM) is explained, where a ROI (Region of Interest) in a image can be accessed with a unique representation space, avoiding additional processing over the Image Space, in order to preserve the quality of desired details. In this case, a Region of Interest can be displayed, with the same perceived-distortion features, but with features calculated from the representation space itself. This method is compared with theoretical ROI access method in order to validate it, making a comparison of the points that the Method has detected, by using a distortion metric over the calculated positions. The choice for wavelet representation is extended not only for the feature extraction of ROI but also for the metric space, in which the rate-distortion optimization plays a role. This translation of the rate-distortion curve on the Wavelet representation prevents the additional processing of coding the whole source image in the source representation space, by using the Second Order Model of Information measure (SOM) in order to calculate the goal entropy for the given distortion measure set. This SOM is widely used, because it considers high correlation features of image between neighbor elements. This method provides a support for the processing of images that are coded in another representation spaces like DCT, reducing the amount of computational load for mandatory new implementations, like edge extraction or texture classification.