A multi-scale arc height shape description(MSAH)is proposed for image retrieval.This method uses the multi-scale arc height to measure the bending degree of each contour point.At each scale

an arch height function is generated and is suffered from Fourier transform.The Low frequency coefficients of Fourier transform for arch height functions of each scale are combined to form a feature vector as the shape description.At the shape matching stage

the L

1

distance between the two features is used to measure the shape similarity.MSAH can not only capture the global features of the shape

but also reflect the details of the shape

therefor it has a higher retrieval

accuracy when applied to image retrieval.MSAH is very compact and has a low computational complexity

so it has higher retrieval efficiency.MSAH is evaluated on the MPEG-7 shape database and plant leaf database

all the experiments show that MSAH has higher retrieval performance over the recent proposed transform domain based methods and shape contexts method.