Digital Image Correlation (DIC) is a robust, non-contact method for full-field deformation measurement widely used in photomechanics. Based on subset image matching and an iterative optimization architecture, traditional DIC deformation field analysis requires manual selection of such calculation parameters as subset size, rendering it challenging to balance spatial resolution and measurement resolution in non-uniform field measurements. The requirement of manual parameter adjustment is lowered by a recently emerged supervised learning-based neural network DIC method, as it establishes a direct correlation between images and deformation fields using deep neural networks (DNN). However, its generalization and accuracy are contingent on the scope and variety of the dataset because it requires training with a large number of simulated speckle images of deformation fields with known standards. This paper proposes an unsupervised learning neural network based DIC measurement method that circumvents the need for standard deformation fields and achieves high precision analysis of deformed fields. It constructs a loss function based on the photometric consistency assumption of speckle images before and after deformation and allows the neural network model to directly extract displacement information from the gray data of the “reference image-deformed image” pair. Meanwhile, the model is enhanced accordingly by further introducing displacement continuity constraints to enhance accuracy and mitigate noise interference. Validation through both simulated and real experiments shows that the proposed approach surpasses traditional subset DIC in measuring non-uniform deformation fields with higher precision. It also offers superior flexibility and adaptability over supervised learning-based neural network DIC methods by eliminating the need for a large training dataset, allowing for immediate application in displacement field analysis.