Because infrared images usually have the disadvantages of low contrast, blurry edge, low signal-to-noise ratio (SNR) and less gray levels, an improved infrared image enhancement algorithm based on histogram equalization is proposed. Firstly, by adaptively selecting an proper threshold, the redundant gray levels which have pixel sum less than the threshold are completely compressed and those effective gray levels which have pixel sum greater than the threshold are stretched on the basis of a very rational logarithm transformation. Secondly, a conventional histogram equalization method is used to process the stretched gray levels. Finally, all these equalized gray levels are arranged at an equal distance in the whole gray level area. To meet the real-time requirement, an infrared image enhancement system which uses a TMS320DM642 chip as its core processor is designed. The experimental result shows that the image enhanced by using this algorithm is obviously better than the one enhanced by using the conventional histogram equalization algorithm in quality. The object-background contrast and definition of the image are both improved and the edge details are kept. Moreover, it takes only 16.73 ms for the algorithm to process an image with $320\times256\times8$ bits. So, the requirement of real-time image processing can be satisfied completely.