As digital images continue to proliferate, content-based image retrieval represents an increasingly formidable task. There is a huge production of digital images in every domain of life. So, every field gains benefits from CBIR. Extensive research has been done to improve CBIR as many minute details of an image such as texture are not properly retrieved. In many images, all the features are not extracted because an image's directional and multiscale information and color features are missing. In this research, we proposed a transform domain system to effectively retrieve an image's texture and basic content. In this method, we first  used the wavelet transform to obtain directional and multiscale information about an image and then fused it with tri-directional pattern to gain confusion matrix of 15 local patterns. Lastly, we combined the patterns to retrieve similar images. We used Corel 5k dataset on natural images. Our research turns out to give 62% precision rate of retrieved images. The research outcomes confirm the method’s efficiency in the retrieval of images.

Keywords: Image retrieval, CBIR, digital images, Corel 5k dataset, multiscale information.