A Novel Entropy-Based Approach for Thermal Image Segmentation Using Multilevel Thresholding

Abstract

Image segmentation is a fundamental challenge in computer vision, transforming complex image representations into meaningful, analyzable components. While entropy-based multilevel thresholding techniques, including Otsu, Shannon, fuzzy, Tsallis, Renyi, and Kapur approaches, have shown potential in image segmentation, they encounter significant limitations when processing thermal images, such as poor spatial resolution, low contrast, lack of color and texture information, and susceptibility to noise and background clutter. This paper introduces a novel adaptive unsupervised entropy algorithm (A-Entropy) to enhance multilevel thresholding for thermal image segmentation. Our key contributions include (i) an image-dependent thermal enhancement technique specifically designed for thermal images to improve visibility and contrast in regions of interest, (ii) a so-called A-Entropy concept for unsupervised thermal image thresholding, and (iii) a comprehensive evaluation using the Benchmarking IR Dataset for Surveillance with Aerial Intelligence (BIRDSAI). Experimental results demonstrate the superiority of our proposal compared to other state-of-the-art methods on the BIRDSAI dataset, which comprises both real and synthetic thermal images with substantial variations in scale, contrast, background clutter, and noise. Comparative analysis indicates improved segmentation accuracy and robustness compared to traditional entropy-based methods. The framework's versatility suggests promising applications in brain tumor detection, optical character recognition, thermal energy leakage detection, and face recognition.

Keywords: entropy; segmentation; thermal images.

Figure 1

Comparison of pixel shuffling in grayscale image with entropy and standard deviation values: (a) original image; (b) image with pixels shuffled row-wise; (c) image with pixels shuffled column-wise; (d) fully shuffled image (rows and columns); (e) image histogram.

Figure 2

Comparison of thermal image enhancement with different Degrees of Enhancement (DoEs): (a) original image; (b) 25% DoE; (c) 50% DoE; (d) 75% DoE; (e) 100% DoE.

Figure 3

Monotonic increase in kernel-based metric values with higher Degrees of Enhancement (DoEs) depicted in Figure 2: (a) original image; (b) 25% DoE; (c) 50% DoE; (d) 75% DoE; (e) 100% DoE.

Figure 4

Transformation functions with different ${\gamma }_{\mathsf{\phi }}$.

Figure 5

Comparison of thermal image with different $\gamma$: (a) input image; (b) $\gamma =1.00$; (c) $\gamma =0.50$; (d) $\gamma =0.25$; (e) $\gamma =0.10$; (f) $\gamma =0.01$; (g) $\gamma =0.001$; (h) $\gamma =e^{\left({\displaystyle \frac{\mu -\max \left\{I\right\}/2}{\max \left\{I\right\}/2}}\right)}$.

Figure 6

Comparative analysis of segmentation accuracy between the proposed model and existing entropy-based functions: (a) input image; (b) Kapur segmentation; (c) Masi segmentation; (d) entropy functions; (e) Renyi segmentation; (f) proposed segmentation.

Figure 7

Comparative analysis of enhancement performance with different (a) input images; (b) enhanced images $\rho =\mu /10\gamma$; (c) histogram of (b); (d) enhanced images $\rho =2$; (e) histogram of (d).

Figure 8

Test images and their histograms for: (a) Image4; (b) Image5; (c) Image6.

Figure 9

Sample images from the real and synthetic datasets. From left to right: small, medium, and large objects. The two images in (a) are real images of animals and humans, respectively, while (b) presents synthetic images of animals and humans. The synthetic data comprise a mixture of summer and winter scenes, with winter scenes featuring dark trees against the ground.

Figure 10

Sample images from the real-world kangaroo (Macropodidae) dataset.

Figure 11

Ground truth: (a) Image4; (b) Image5; (c) Image6.

Figure 12

Thresholding results for Image4: (a) Shannon [48]; (b) Tsallis [42]; (c) Renyi [43]; (d) Kapur [4]; (e) Masi [44]; (f) proposed method (k = 1, t = 45; k = 2, t = 94, 95; and k = 3, t = 123, 124, 125).

Figure 12

Thresholding results for Image4: (a) Shannon [48]; (b) Tsallis [42]; (c) Renyi [43]; (d) Kapur [4]; (e) Masi [44]; (f) proposed method (k = 1, t = 45; k = 2, t = 94, 95; and k = 3, t = 123, 124, 125).

Figure 13

Thresholding results for Image5: (a) Shannon [48]; (b) Tsallis [42]; (c) Renyi [43]; (d) Kapur [4]; (e) Masi [44]; (f) proposed method (k = 1, t = 50; k = 2, t = 98, 99; and k = 3, t = 128, 129, 130).

Figure 13

Thresholding results for Image5: (a) Shannon [48]; (b) Tsallis [42]; (c) Renyi [43]; (d) Kapur [4]; (e) Masi [44]; (f) proposed method (k = 1, t = 50; k = 2, t = 98, 99; and k = 3, t = 128, 129, 130).

Figure 14

Thresholding results for Image6: (a) Shannon [48]; (b) Tsallis [42]; (c) Renyi [43]; (d) Kapur [4]; (e) Masi [44]; (f) proposed method (k = 1, t = 47; k = 2, t = 97, 98; and k = 3, t = 127, 128, 129).

Figure 15

Comparison of a thermal image: (a) original, (b) single thresholding, and (c) multilevel thresholding.