A new texture-based anomaly detection(TBAD) approach was presented,which segmented one image into different textures and analyzes the distribution of pixel values of the textures.TBAD assumes that the background pixel values within textures can be modeled as Gaussian distributions with mean values that vary texture-to-texture.And the anomalies(man-made objects) have values that deviate significantly from the distribution of the texture.TBAD estimates background statistics over segmented textures,so it can detect objects of any size or shape.Extensive experiments applied to the real images of small target and extend target validate the good performance of the approach.