Abstract: A leaf area has been one of the most important indicators of photosynthesis, transpiration, respiration, and yield components of plants. The physiological and ecological indicators can dominate plant growth, fruit development, and quality formation. The purpose of this study is to measure the leaf area of cotton by the thermal infrared and visible images. An accurate, convenient, stable, and nondestructive approach was also proposed for the early leaf area measurement in physiological and ecological research. The experimental cotton was cultivated in the greenhouse of the East Field Experimental Base of Cotton Research Institute, Chinese Academy of Agricultural Sciences from July to September 2021 (On July 20th, the cotton seeds were soaked in the hydrogen peroxide for two hours, and then sown in the pots, one cotton seedling per pot, totally 15 pots). When the cotton was in the seedling stage (August 25th, the number of euphylla was 1-4), the infrared imager T660 was used to take photos at 14:00 pm, where the radiation difference among soil, leaf and shadow reached the outstanding effect. Five pots of cotton seedlings were randomly selected to capture the images. Both thermal infrared and visible images were obtained eight from each pot. Taken together, 16 images were obtained from each cotton pot. A hand-held standard board with a circular reference was used to hold the inclined leaves in each capture, in order to reduce the distortion of cotton leaves in the image. Hough circle detection was used to extract the region of reference substance in the visible image. The GrabCut was used to extract the leaf regions in the thermal infrared image. The capture and thermal infrared images were firstly adjusted by the FLIR tools. After that, the pixel values of leaf regions of the two images were assigned the weights and then superimposed. The color filling was carried out using a 4-connected field, in order to eliminate the isolated pixels near the leaf regions. The pixel value of the leaf area was defined after the color filled the connected area. The leaf regions were converted into white (pixel value is 1, 1, 1), whereas, the rest was converted into black (0, 0, 0), according to the pixel values of the leaves. The following step was to convert the 3-channel image with the leaf information into a single-channel image. Then, the contour was extracted from the reference substance and the leaf regions. The leaf area was then calculated, according to the multiple relationships of the number of the pixels. The study-cutting weighing and Image Pro Plus image were used to measure the five pots of cotton seedlings for eight times after capture. The correlation analysis showed that there was a significant linear correlation (r>0.99, P<0.01). The difference between the method proposed in this paper, weighing, and Image Pro Plus method are all in the 0.06%-6.73% range. Additionally, the higher stability of the measurement was achieved, where the average coefficient of variation was 0.78%. Therefore, an accurate, stable, rapid, and nondestructive method can provide a promising convenience for physiological and ecological research in the early leaf area measurement.