![]() | Only 14 pages are availabe for public view |
Abstract This thesis is concerned with the processing of infrared (IR) images. The objective is to enhance the quality of these images as they have low contrast and low SNR. Generally, the IR images are acquired with thermal cameras that can take still images or video sequences. These cameras record the thermal distribution of the object of interest and the background. Different sources of heat may lead to noise in the obtained images and may lead also to low contrast around the object of interest. A non-uniformity effect appears in the background of IR video frames with the heating of the imaging camera or system. This effect appears as a fixed pattern noise in the background. This fixed pattern noise leads to drift of the IR image histogram to the right on the gray-scale axis. This problem is treated in this thesis with a Histogram Matching (HM) treatment. The objective of this treatment is to correct the subsequent frames in a video sequence based on the histogram of the first frame in the sequence that does not suffer from the heating effect. That is why we adopt the HM as an efficient tool for Non-Uniformity Correction (NUC) in IR video sequences. In addition, this thesis presents an IR image enhancement scheme that addresses the low-contrast problem depending on Contrast Limited Adaptive Histogram Equalization (CLAHE) and fuzzy logic. The objective of this scheme is to enhance the visual quality of IR images leading to better ability to discriminate the object from the background. A new framework is proposed in this thesis for efficient object detection from IR images. The IR image is first enhanced. After that, the image gradient is obtained with a Laplacian operator. After that, the histogram of gradients is estimated, and finally the cumulative histogram is obtained. The cumulative histograms for images with and without objects are used for object detection. Certain bins from these cumulative histograms are selected for detection based on estimated thresholds. Simulation results show the success of the proposed techniques for NUC, IR image enhancement and object detection from IR images. |