Accurately and sensitively detecting enhancer-promoter loops from genome-wide interaction matrices, such as Hi-C data, is critical for advancing our understanding of gene regulation. Normalization methods like Iterative Correction and Eigenvector decomposition (ICE) have been widely applied before chromatin loop detection to remove potential biases in Hi-C data. However, we found that ICE and similar methods tend to specifically strengthen the signals of structural or CTCF-associated loops while weakening enhancer-promoter interaction signals, making accurate detection of enhancer-promoter loops more challenging. To address this issue, we developed Raichu, a novel method for normalizing chromatin contact data. Raichu detects nearly twice as many chromatin loops as ICE from Hi-C data. It not only recovers nearly all loops detected by ICE but also identifies over ten thousand enhancer-promoter loops that ICE fails to detect. With its improved power in detecting these regulatory loops, Raichu identifies more biologically meaningful differential loops between conditions in the same cell types. Additionally, we show that Raichu performs well across different sequencing depths and platforms, highlighting its potential to provide novel insights into 3D genomic organization and transcriptional regulation.