Recently, retrieving and searching for relevant images from large datasets has become increasingly challenging for researchers. Hashing methods that mapped raw data into a short binary code have attracted increasing attention from the researcher. Most existing hashing approaches map samples to a binary vector with a single linear projection. This paper introduces a CNN-based hashing method that uses multiple nonlinear projections to produce additional short-bit binary code. Applying a convolutional neural network, we obtain an end-to-end. In order to maintain similarity between images, a loss function was also designed. Moreover, it minimized the quantization error by providing a uniform distribution of the hash bits. Extensive experimental results from various image retrieval benchmarks show the superiority of the proposed method and demonstrate its efficient performance compared with other state-of-art methods.