Abstract: In the machine vision-based intelligent system for recognizing and sorting male or female silkworm pupa, the quality of silkworm pupa images is the key for accurate recognition. Low illumination and noise，as the main factors degrading silkworm pupa images, can give rise to the loss of images textures and structures to a great extent, which brings a challenge for intelligent system to identify silkworm pupa's gender. State-of-the-art methods, like Shan's work, can be ineffective when images are perturbed by noise. The main contribution of our work was the effective elimination of noise by Tikhonov regularization while restoring image contrast and preserving image textures and structures based on the Shan's modeling. In order to improve the quality of degraded silkworm pupa images, a novel method combining tone mapping with Tikhonov regularization, which was capable of enhancing image contrast and compressing noise simultaneously, was proposed in this paper. According to Shan's work, it was assumed that Low-illumination image is obtained via dynamically compressing a Low-illumination image. A 3×3 neighborhood of pixels in both images was defined. The linear functions mapping locally radiance in such a 3×3 window of low-illumination silkworm pupa images to that of the desired ones were formulated. The monotonicity of linear functions can preserve local structural information of image. Through integrating these linear functions, the image-level objective function was established to restore image illumination contrast. Further, Tikhonov regularization implemented by the Laplace of Gaussian (LoG) operator was used to obtain the final objective function. Tikhonov regularization could not only smooth noise and preserve structural information of image effectively but also could be beneficial to a stable solution in the iterative processes. The existence and unique of solution of the objective function was addressed via verification of convexity according to D. P. Bertsekas' convex theory. After coefficients of local functions aforementioned were solved analytically, the global optimal restored silkworm pupa image was obtained by linearly optimizing objective function. In the experimental section, simulated data and real data (both including male and female silkworm pupa images) experiments were conducted on the platform configured with CPU i5, 2.4G Hz, memory 2G, and 32 bit operation system and matlab2012. The result showed that the performance of the proposed method was better than Shan's method, especially, when quality of images was degraded by noise and low illumination at the same time. Noise can be greatly removed while the image contrast was improved and the details were preserved. Moreover, the proposed method can be conveniently extend to improve the quality of low-illumination plant images, such as vegetables, because only one parameter, i.e., regularization parameter, needed to adjust by trial and error until the best results were obtained in the implementation of the proposed method while other parameters involved basically were the constant values. The automatic selection of regularization parameter can be achieved by cross validation of L curve and U curve. The proposed method can benefit a wide application of machine vision technologies, such as biological measurement and pattern recognition, in agricultural fields.