A robust content based image retrieval using local full-directional pattern (LFDP)
Behzad Merhrbakhsh Choobari, Saeed Mozaffari · 2017
In this paper; we propose new method named local full-directional pattern (LFDP) for content-based image retrieval (CBIR). In addition, instead of applying the algorithm to the image itself, we apply it to a new image constructed by getting mean of 3×3 sub-regions gray value as each pixel's value. In local binary patter (LBP) the gray value difference of the central pixel and its neighboring pixels is calculated. In the proposed method, eight distinct directions are defined. Each pixel gets one of the directions according to its relationship with its surrounding neighbors, which is calculated using first-order derivatives in vertical, horizontal and two diagonal directions. We used Corel 1000 database to compare our method with LBP and local tetra pattern (LTrP). Our proposed method shows significant progress in both average precision and recall. Due to applying the algorithm to the mean value of 9 pixel-windows, the proposed method shows better retrieval results for noisy images.