Abstract: here is not a mature algorithm for density recognition of sweater texture at present. This paper carries out reseach with this problem, and an automatic recognition method based on extreme cluster is proposed. Firstly, actual size of any recognition areas is calculated, image is made enhancement processing with linear method. Then a fusion algorithm of extreme is proposed. With this algorithm, gray peak cluster and valley cluster generated by loop column and void part in image are extracted, same extreme clusters in same position are fused, and fusion graph of gray extreme cluster for image change law is drew. Finially, calculation formula of lateral and longitudinal density of sweater texture is given. The algorithm in this paper is verificated by Matlab 7.0 and result is concluded that recognition accuracy is more than 98% for jersey and rib texture.