Single-pixel imaging (SPI) captures two-dimensional (2D) images by using spatial light modulators and measuring the light intensity signal, which has emerged as a promising technology. It can be achieved through single-pixel camera schemes using structured detection or ghost imaging schemes using structured illumination. Over the past decade, there have been many developments for x-ray imaging [1,2], microscope [3,4], three-dimensional imaging [[5], [6], [7], [8]], object tracking [9,10] and encryption [11].

The image quality of SPI is restricted by the number of patterns. Mask patterns can be divided into random patterns [12], Hadamard basis patterns [13], and Fourier basis patterns [14]. Although patterns with orthogonality have better performance, random patterns are still chosen as modulation patterns in recent studies [7,[15], [16], [17]]. The optimization of SPI using random patterns can be achieved through structural modification strategies, such as hybrid mode [18], foveated structure [19] and deep-learning basis [20,21]. Results with higher quality can be obtained by ensuring that local feature areas in images are properly modulated by patterns optimized in certain areas. Random patterns for modulation are specifically crafted in these methods. Moreover, only the displayed structure of modulation patterns is used in the SPI reconstructions mentioned above. However, each modulation pattern is not merely composed of displayed structural information. The hidden pattern information (HPI) that naturally exists in patterns is ignored in SPI reconstructions.

Here, we propose a SPI reconstruction strategy enhanced by HPI in random patterns, which does not require modifying modulation patterns. We employ the equivalency of structures in random binary patterns to generate extra pattern information in the image reconstruction stage. We utilize a simple algorithm to obtain resulting images from multiscale information. We experimentally demonstrate that more accurate structural information is recovered by using added HPI in the same measurement conditions.