Due to complex and changeful environment, the image segmentation of fruits with diseases in natural scenes is a difficult problem. A logarithmic similarity constraint Otsu and level set active contour (LSAC) based image segmentation approach of fruits with diseases was proposed in this paper. Considering the complexity and changeableness in natural scenes, the constraint Otsu method for segmenting logarithmic similarity image between diseased fruits and samples was introduced to distinguish diseased fruits and background; because of the local optimality of LSAC, improved distance regularization level set evolution (DRLSE) with adaptive expansion coefficient was used to lead contour to actual position. Firstly, the sample color of fruits with diseases, which included not only health area but also diseases area, was modeled using Gaussian mixture model (GMM), and then the logarithmic similarity between the image of fruits with diseases and model was obtained. Secondly, logarithmic similarity image was segmented with constraint Otsu and then morphology operator was used to filter out noise and interference. Thirdly, leastsquares ellipse fitting method was employed to further removal interference and get initial contour for LSAC. Finally, the contour of fruits with diseases was evolved to the actual position taking use of improved DRLSE with adaptive expansion coefficient. The experimental results show that the actual contour of fruits with disease in complex natural scenes can be obtained and the proposed method can provide the basis for the subsequent diseases density estimation and prevention of fruit diseases.