In the realm of image retrieval, the efficient management of images has become increasingly intricate, leading researchers to explore diverse texture features such as characteristics determined by edges, directional qualities, rotation invariance, and homogeneity. Nevertheless, modern approaches frequently convert the boundary-to-center correlation into a local pattern, which is then represented as a feature vector using histograms. This paper tackles the finding and retrieving of images from massive storage systems. The proposed system introduces a image retrieval method known as Local Extrema Co-occurrence Patterns (LECoP) which uses the HSV color space. It extracts color, brightness, and intensity information from photographs. Gray-level co-occurrence matrix (GLCM) is used to capture co-occurrence associations among pixels in the Local Extrema Pattern (LEP) map. LEPs define local details within an image. The gray-level co-occurrence matrix efficiently extracts orientation data from the LEP, converting it into a refined feature vector. This paper delves into a parallelized image feature extraction method using Compute Unified Device Architecture (CUDA), comparing its performance against a sequential approach. The results highlight the substantial efficiency gains achieved through parallelization, demonstrating its potential to reduce overall computation time in image feature extraction significantly.