The present invention discloses an improved super-pixel-based image significance detection method. The method comprises the following steps: firstly, segmenting an original image into super-pixels consistent in color and texture based on the simple and linear iterative clustering super pixel segmentation algorithm; secondly, for the image super-pixel segmentation result, calculating the initial saliency map of the original image based on the sparse representation theory; thirdly, for the image super-pixel segmentation result, calculating the center-edge weight map of the original image based on the center-edge idea; fourthly, for the image super-pixel segmentation result, clustering the super-pixels based on the normalized theory so as to obtain a plurality of clustering areas; fifthly, based on the above result, calculating the final saliency map of the original image. According to the technical scheme of the invention, compared with the traditional super-pixel-based image saliency detection method, the problems, including the fuzzy boundary of a saliency object, the inhibited interior of the saliency object and the like, can be solved. The saliency object can be highlighted more uniformly. Meanwhile, the background is effectively suppressed, so that a better detection result is obtained.