Combining genetic algorithm with human visual apperception theory , a new multi-threshold image segmentation in complex background , i. e. an method of image threshold segmentation based on genetic algorithms , is discussed , which can distinguish changeable target backgrounds automatically and regulate the best segmenting region. Much more accurate results of segmentation and robustness are obtained which helps overcome drawbacks of traditional threshold methods. Computer emulation experiment shows that the performance of the algorithm provided above are satisfactory.