Evaluation method for inter-chip uniformity in infrared mosaic arrays

Infrared mosaic arrays are an effective strategy for improving detection resolution in aerospace applications. This Letter proposes a method for effectively evaluating the inter-chip uniformity of infrared mosaic array images. Using human visual perception scores as reference labels, three core metrics-neighborhood difference (ND), Laplacian response (LR), and local standard deviation (LSTD)-are selected due to their high sensitivity to local variations and fused using a support vector regression (SVR) model. Leave-one-out cross-validation (LOOCV) demonstrates that the proposed evaluation method exhibits satisfactory generalization capability on unseen mosaic configurations. An ablation study indicates that the three metrics are complementary and provide stable performance in characterizing inter-chip uniformity. Furthermore, validation on additional independent datasets confirms the robustness of the proposed method. The proposed evaluation method can guide the selection among candidate stitching sequences and can also be employed to assess the inter-chip correction performance of non-uniformity correction in infrared mosaic arrays.