The content-based image retrieval (CBIR) has been applied in the image processing as well as pattern recognition. A challenging task in the CBIR research has been through the feature extraction for decreasing any semantic gap that is an active research topic. Here, in this work, there is a texture feature that is extracted from an image, making use of a technique of curvelet transform. This curvelet is selected for a sparse representation that is quite critical for the estimation of images, which have been de-noised with some inversed problems. The wrapping-based curvelet transform will be even more robust as well as faster in the time of computation than the Ridge Transform. A technique of feature selection will be brought for selecting the optimal features. The correlation-based feature selection (CFS) method has been adapted for improving the accuracy of the CBIR systems. The non-deterministic polynomial (NP)-hard problems may be solved using the chemical reaction optimization (CRO) that has been motivated using the technique of chemical reaction. The benefits of proposed method is a necessity for marginal human intrusion for the purpose of retrieving the images needed from its database. The proposed mechanism has been again assessed depending on the Coral Database, and a performance study is ended using precision, f measure, and recall.