High-resolution infrared array detectors are prohibitively expensive and technologically challenging to manufacture. Block-based compressive imaging (BCI) provides an alternative solution for high-resolution infrared imaging. However, the quality of object reconstructions in BCI is often degraded by block artifacts, which is inherited in the method. This is particularly important for a small block size, which is common in BCI, associated with a limited demagnification factor from a spatial light modulator (SLM) to a detector array. To address the issue, in this work, we propose a deep unfolding network SBCI-DUN, which introduces a proximal mapping module that incorporates a deblocking module (DM) and a deformable convolution (DCN). DM is crucial to mitigating block artifacts. DCN enhances the ability of the model to capture fine details and establish long-range dependencies through flexible local modeling and adaptive feature extraction while maintaining relatively low computational cost. Extensive evaluations on simulated datasets and real-world near-infrared target data demonstrate that SBCI-DUN outperforms existing networks in reconstruction quality.