Efficient medical image retrieval system using Geometric Invariant Point Bilateral Transformation (GIPBT)

Abstract

Sensor rich environments assist CBIR system to provide granularity in detail for many applications. From sensor rich environments, CBIR can provide the process to exploit images that can integrate both search & retrieval mechanisms. Era of digital technology with Smart devices connected to internet so as to communicate is made possible with Internet of Things (IoT). As IoT deals with huge data in the form of images or videos, to be communicated between smart devices through IoT, deals with complex data processing. Proposed Image Retrieval System reduces the computation overhead for wide range of IoT applications. Medical Images are not only image which holds the pixel also the valuable information of medical diagnosis of every human being. Medical image retrieval has become an inescapable task in every hospital IT system due to the influx of medical pictures. A novel texture fusion technique and a rapid similarity measure metric were used in this study to create an efficient medical CBIR. Proposed feature extraction method of Geometric Invariant Point Bilateral Transformation (GIPBT) algorithm provides the texture information. In this method features are computed as point-to-point geometric information of bilateral filtered image. It automatically inoculates the universal information into the local descriptor and reinforces the feature depiction. To further enhance the descriptor, we used Histogram of Second Order Gradient (HSOG) feature extraction method. By use of these two features, we are performing the feature matching. To analyze and categorize the non-linearity between the features, we propose the kernalized discriminant analysis concept into the score prediction metric. Our proposed similarity measure method is Discriminant Kernalized Disparity (DKD) metric. In DKD, kernel applied Eigen decomposition is utilized for each feature set and disparity metric is evaluated. Finally the proposed system is simulated and evaluated with different medical database images and retrieval performance of Retrieval Efficiency, Normalized Average Rank of Retrieval, Precision, Recall, SSIM and processing Time. Some previous implementations are examined and compared to our results. The proposed method has produced Retrieval efficiency of 98.4%, NARR of 15.18%, Precision and Recall of 98.4%, F1 score of 98.39%, Correlation of 98.2%, SSIM of 97.92% and all at a speed of 30 s 5% more precision as well as less processing time are achieved compared to prior efforts by our proposed solution.