<p>Harris Hawks Optimisation (HHO), as a typical metaheuristic algorithm, can effectively reduce computational workload and improve optimisation efficiency. It is currently widely used for multi-level threshold segmentation in image processing. However, in this algorithm, the simple random population initialisation makes it unstable. And there is an imbalance in the exploration and evolution capabilities of the algorithms. It is easy to fall into a local optimal state. At the same time, the lack of a mutation mechanism in the standard Harris-Hawk algorithm leads to a decrease in population diversity in the later stages of algorithm iteration, making it easy to fall into local optima. To address these issues, this paper proposes a new HHO threshold segmentation algorithm. The proposed method uses logistic chaotic mapping to increase the diversity of solutions. In the follower position update strategy, the Cauchy mutation operator and the opposition-based learning (OBL) strategy are introduced to perturb the optimal solution position, thereby generating new solutions and enhancing the algorithm’s ability to escape from local space. The introduction of Gaussian mutation serves to improve the original single Levy flight, strengthen the distribution of the population, and improve the global search ability. Eight metrics, including PSNR and SSIM, were used to perform comparative experiments on six benchmark images and eight images from the Berkeley BSD500 datasets. Compared to existing methods, the results obtained illustrate the superiority of the proposed approach. In order to demonstrate the effectiveness of the proposed algorithm, we conducted comparative experiments with other algorithms on the CEC 2020 benchmark function test suite. The experimental results verified by Friedman test statistics show that the proposed algorithm ranks first compared to other algorithms, with a Friedman mean rank of 1.6, and produces efficient and reliable results in terms of both consistency and accuracy.</p>