Parallelization of Local Diagonal Extrema Pattern Using a Graphical Processing Unit and Its Optimization

The incorporation of medical imaging devices in diagnosis has resulted in huge collection of medical images in hospitals and health centres. A search for a similar image from this image collection corresponding to a new medical image is a much needed help for junior doctors or students. This task involves describing each image in the image collection and also new image. After this, a similarity measure is applied between the image in the image collection and new image. Texture features have been found to be efficient in describing medical images owing to their high discerning capability. Various texture features have been introduced by researchers. Local Diagonal Extrema Pattern (LDEP) is a texture feature that uses only local diagonal neighbours, and hence the dimensionality of resulting feature vector is reduced. In this chapter we discuss parallel LDEP extractor on a GPU using CUDA. A constant kernel execution time for medical images of various sizes has been obtained on a GeForce GTX 1050 GPU.