the existence of haze greatly degrades the image quality and hence decreases the cognition performance of a vision system. therefore, it is crucial to remove haze from images. instead of formulating dehazing as an image rest-oration or mathematical inversion problem, we, in this paper, conduct dehazing by learning a proper transformation function (i. e., enhancement gain) under the framework of classical image enhancement. there are three novelties. (1) it is observed that intensity-inverted hazy (foggy) image and low-light (i. e., underexposed, low-dynamic range) image are similar in the sense of properties of dark color and low-dynamic range. based on this observation, it is straightforward to invert the intensity and then utilize low-lightoriented tone mapping in large-scale image layer to remove haze from a single hazy image. however, this simple intensity inverting plus tone mapping does not directly result in satisfying dehazing effect. (2) to make the inversion plus mapping method work, we propose an intensity smoothing algorithm consisting of maximum-based blocking and bilateral filtering, which results in remarkable dehazing result. (3) an algorithm is proposed to learn optimal tone mapping. though our method does not rely on the imaging model of hazy image, experimental results demonstrate that our enhancement method is better than the model-based methods such as dark channel prior and its variants. the proposed method is called iitem. one key of item is intensity inverting and the other key is learning-based tone mapping. by learning, the tone mapping is optimal in the sense of haze removal.