COVID-19 X-ray images are a vital approach for diagnosing whether a patient has an infection. By using multi-threshold image segmentation (MIS) technology to segment the target area of the COVID-19 X-ray image automatically, doctors can more efficiently determine whether the patient is infected with the virus and the current course of the disease. Nevertheless, as the threshold value rises, the computing cost of MIS approaches grows considerably, and for this reason, many researchers utilize meta-heuristic algorithms (MAs) as optimizers to select the optimal thresholds. Yet some issues cause slow convergence and local optimum solutions stalling. To revise the drawbacks, this paper proposes a strengthened version of the hunger games search (HGS), titled CDHGS. CDHGS introduces crisscross optimizer (CSO) and dimension learning-based hunting (DLH) mechanisms to HGS. First, CSO allows different individuals to exchange information, which speeds up convergence. Then, DLH mines more details on an individual's surrounding neighbors, thus alleviating the local optimum problem of the algorithm. A series of comparative experiments completed at CEC2014 showed that the proposed CDHGS has superior performance in respect of optimization than other advanced algorithms. Besides, a CDHGS-based MIS method is presented and employed to segment COVID-19 X-ray images. Specifically, we build a two-dimensional (2D) histogram utilizing non-local mean and grayscale images to illustrate the information of images, use Rényi's entropy as the objective function, and maximize Rényi's entropy to find the optimal thresholds. The COVID-19 X-ray image segmentation (IS) results of the evaluation display that the CDHGS-based MIS can obtain considerably exceptional segmentation results and stronger robustness than other segmentation methods. In all, CDHGS is a competitive approach in both global optimization (GO) and IS.