This paper aims to effectively overcome the problem of low prediction accuracy and unstable output caused by the random selection of weight threshold of BP neural network algorithm. The BCC-BP neural network algorithm combining the bacterial colony chemotaxis (BCC) algorithm and BP neural network algorithm was proposed and the algorithm was applied to RGB to LAB color space conversion model. The BCC algorithm was used to select the initial weight threshold of the BP neural network to overcome the problems caused by the random selection of the initial weights and thresholds. According to the requirement that the allowable error range of national ordinary printed matter is below 6 standard chromatic aberration units, the prediction accuracy of BCC-BP algorithm is 81.07% when the chromatic aberration is less than 6, which is better than BP, GA-BP and PSO-BP algorithms. At the same time, the average chromatic aberration ΔE is less than 6 standard chromatic aberration units. The prediction results of BCC-BP algorithm for 10 times are all lower than 6. BCC algorithm is used to assist BP neural network to select the initial weights and thresholds, which can effectively improve the output accuracy and stability of the BP neural network model in the application of color space conversion.