The invention provides a
single image haze removing method based on a mean-
mean square error dark channel under a superpixel framework. The method comprises the following steps that the minimum value matrix I<dark> of each color channel of an image I is calculated according to the formula (6), hereafter referred to as the
grayscale matrix I<dark>; appropriate parameters T and k are obtained according to the step 1, and the
haze removing rate
omega is calculated according to the formula (5); the atmospheric light value A is estimated; the matrix
img=I / A is calculated, and
superpixel segmentation is performed on the matrix
img so that multiple
omega of which the imaging field depth d(x) of the scene and the
scattering coefficient beta of the atmospheric medium are constant are obtained; the dark channel is calculated for each
omega according to the formula (4) so as to obtain J<dark>, and the obtained J<dark> of all the omega is spliced together so as to obtain the dark channel J<dark> of the whole image; the coarse
transmittance t is calculated through t=1-omega*J<dark>; fine treatment is performed on the coarse
transmittance t so as to obtain the fined
transmittance t*; and the final recovered image J is obtained by the formula J=(I-A) / t*+A. According to the method, the
haze concentration and the field depth are maintained to be unchanged in the local area represented by the superpixels so that the
halo effect of the abrupt change part of the field depth can be overcome, and the problem of color cast of the infinite distance part of the field depth can be effectively alleviated.