The existing evidence collection techniques for fingerprints and traces are usually invasive. Consequently, the original evidence in the crime scene tends to be damaged. As one of the most important materials for detecting and searching crime scene evidence, forensic image has become a research focus in recent years. With the development of technology, criminals have a better understanding of the general principle and mechanism of forensic images. Criminals are likeyl to use various methods to avoid traditional forensic image technologies, such as wrapping their hands with gloves and plastic bags to eliminate the hand traces. These methods, make it hard to obtain valid information via traditional forensic images, which means valid information can not be gained.

Faced with these challenges, infrared image is becoming an effective means for obtaining crime scene traces. Even criminals use various methods to rid the traces left at the crime scene, the infrared image is still able to get enough valid information. However, for these crime scene marks destroyed by criminals, the infrared images we are able to obtainare extremely blurred infrared images compared to ordinary infrared image. The target areas of these depth blurred images are seriously affected in two cases. The target area marks is weakened in the first case. Traces of the target area being weakened, and the difference between them and traces of the background areas is no longer obvious. Because the traces intensity of the background area maybe stronger than the target area which is extremely weakened, therefore, we call such images as target weakened traces infrared images. Target area is severely damaged in another case, such as wiping, or external influence. It is difficult for target area of such images to reflect the real target outline, therefore these images are called target drift traces infrared images. Therefore, segmentation of target weakened traces infrared images and target drift traces infrared images is a new theoretical and application problem for image processing and criminal Investigation.

There are many segmentation methods for different kinds of images^{[1, 2]}. Akata Z proposes to view attribute-based image classification as a label-embedding problem: each class is embedded in the space of attribute vectors. Gharipour A proposes an integrated framework consisting of a new level sets based segmentation algorithm and a touching-cell splitting method to segment cell nuclei in fluorescence microscopy images^[3]. Girshick R gives a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 50 percent relative to the previous best result on VOC 2012-achieving a mAP of 62.4 percent^[4]. Christodoulidis A proposes a new hybrid method for the segmentation of the smallest vessels, Line detection and perceptual organization techniques are combined in a multi-scale scheme^[5]. Shi YH presents a novel semi-automatic learning-based prostate segmentation method, for segmenting the prostate in a certain treatment image, the radiation oncologist will be first asked to take a few seconds to manually specify the first and last slices of the prostate^[6]. Zanetti M addresses the numerical minimization of a second-order functional for image segmentation, which exploits a compact matricial formulation of the objective functional and its decomposition into quadratic sparse convex sub-problems^[7]. Kim S formulates a hypergraph-based image segmentation framework in a supervised manner for many high-level computer vision tasks, to consider short-and long-range dependency among various regions of an image and also to incorporate wider selection of features, a higher-order correlation clustering (HO-CC) is incorporated in the framework^[8]. Akselrod-Ballin A presents a novel approach for whole body segmentation of small animals in a multimodal setting of MR, CT and optical imaging. The algorithm integrates multiple imaging sequences into a machine learning framework, which generates supervoxels by an efficient hierarchical agglomerative strategy and utilizes multiple SVM-kNN classifiers each constrained by a heatmap prior region to compose the segmentation^[9].Kumar MP considers the problem of parameter estimation and energy minimization for a region-based semantic segmentation model, and addresss this problem by designing an accurate linear programming based approach for selecting the best set of regions from a large dictionary^[10].

In past years, methods based on minimum distance have attracted more and more attention. Zhu ZC proposes a blind modulation classifier that differs from most existing classifiers, a low complexity minimum distance centroid estimator is suggested to estimate the channel gain and carrier phase jointly^[11]. Bett SK Monitor urbans sprawl using minimum distance supervised classification algorithm in Rustenburg, South Africa, the algorithm is used to classify four land cover classes and post-classification method adopted for change detection^[12]. Kapetanovic D numerically optimizes the multiple-input multiple-output (MIMO) precoders by making use of a state-of-the-art technique for optimization over the unitary group^[13]. Ghimire S introduces a new method of image pixel classification based on minimum distance and hypothesis testing, the method is a nonparametric classification method which uses combined evidence from the multiple hypothesis testings and minimum distance to carry out the classification^[14]. Nasibov E proposes two kinds of simple minimum distance-based classifier methods, these methods and known K-nearest neighbor (KNN) classification algorithm have been performed in order to classify enzymes according to their amino acid composition^[15]. Srivastava MS presents a classification rule based on the minimum distance to classify an observation vector into two or several groups, and introduces a sample-squared distance between the two groups using Moore-Penrose inverse^[16]. Silva WB assesses a new classifier for Polarimetric SAR (PolSAR) images, which input consists of segments, and each one is assigned the class which minimizes a stochastic distance^[17]. Chen YC puts forwards distance-based margin support vector machine for classification, which gives the use of modified slack variables within the SVM (MS-SVM) to solve complex data problems, including class imbalance and overlapping^[18]. Kahraman HT explains and extends a powerful similarity measurement method, which is called the fuzzy distance metric, to measure the distances between the test and training observations^[19].

These algorithms have good segmentation results for different kinds of images, but they are not apply to the target weakened traces infrared images, the first case of the depth blurred infrared images, and cannot effectively solve a variety of criminals’ escape means. Thus, in opposition to the target weakened traces infrared images, we propose a new immune algorithm with minimum mean distance immune field for image segmentation. According to the distinguishing feature of target weakened traces infrared images, this method use multi-step classification algorithm to achieve optimal classification based on the overall statistical properties of target areas and background areas. Through the above steps, this algorithm realizes efficient image segmentation for target weakened trace infrared images.

In order to maintain the physical health, immune factors of biological immunity should identify its own cells and the outside world invading viruses or harmful bacteria, and kill the harmful bacteria and viruses. However, with long-term evolution, viruses and harmful bacteria can evade most of the innate immunity, thus harmful viruses and bacteria can invade the body. The long-term evolution viruses and harmful bacteria already have a great deal of resistance for immune effect molecule in innate immunity. To address the intrusion problem of long-term evolution viruses and bacteria invade, immune system gradually evolves adaptive immunity based on the innate immunity.

Although the long-term evolution viruses and harmful bacteria can evade extermination of innate immune effect molecule, they must exist certain features or some integrated features, which are consistent with the non-evolution bacteria and viruses. Thus, in close collaboration with the innate immunity, the adaptive immune can selection, extraction, and integration the same features between invasion harmful bacteria, viruses and its own cells. If the immune system trains and recognizes these features from a global statistical point of view, it will be able to guarantee the health.

In our previous study, we have given preliminary immune algorithms, but we have not describe immune variation and adaptive immune minimum mean distance recognition in the network and algorithm^{[20, 21, 22]}. In the present paper, we construct a new immune segmentation algorithm with minimum mean distance immune field. Fig.1 shows the whole flow chart of our algorithm.

Fig.1

	Figure Option ViewDownloadNew Window
	Fig.1 New immune segmentation algorithm

When criminals wrap their hands with plastic bags, the infrared image will become target weakened traces infrared image, and the vast majority of criminal investigation means cannot obtain valid information. However, this method is increasingly being used by criminals.

Under laboratory conditions, we shot target weakened traces infrared image when a hand is wrapped with a plastic bag, as shown in Fig.2(a), and we shot hand trace infrared images when the hand leaves for 1 second, 30 seconds and 1 minute, as shown in Fig.2(b), Fig.2(c) and Fig.2(d), respectively. Fig.2(b) can reflect the touching contour between the hand wrapped by a plastic bag and the wall. But Fig.2(b) usually can not be obtained in crime scene because we can not get there in one second. The image we often get is Fig.2(d), which is a target weakened traces infrared image and its gray level can not reflect the valid information.

Fig.2

	Figure Option ViewDownloadNew Window
	Fig.2 Target weakened traces infrared images of hand wrapped by a plastic bag (a): The hand touching the wall; (b): The hand leaves for 1 second; (c): The hand leaves for 30 seconds; (d): The hand leaves for 1 minute

The proposed immune algorithm with minimum mean distance immune field has been simulated in MATLAB program environment for target extraction from target weakened traces infrared image.

Fig.3 shows the extraction results of different algorithms for banded blurred hand trace infrared image. (a) is the hand leaves the gun for 1 second and for 1 minute. (b) is the extraction result of Roberts. (c) is the extraction result of Prewitt. (d) is the extraction result of Sobel. (e) is the extraction result of Log. (f) is the extraction result of Canny. (g) is the extraction result of region growing algorithm. (h) is the extraction result of watershed algorithm. (i) is the extraction result of immune algorithm with minimum mean distance immune field.

Fig.3

	Figure Option ViewDownloadNew Window
	Fig.3 The extraction results (a): Original image; (b): Roberts; (c): Prewitt; (d): Sobel; (e): Log; (f): Canny; (g): Region growing algorithm; (h): Watershed algorithm; (i): Immune algorithm with minimum mean distance immune field

From the segmentation results, we can seen that in these cases with target weakened traces, roberts, sobel, prewitt, Log, canny and region growing algorithm perform poorly, and can not extract hand area from the image. Watershed algorithm performs better but it still produces inaccurate hand boundary contours, and the outline of the five fingers, especially the little finger is very unclear, inaccurate. On the contrary, immune algorithm with minimum mean distance immune field proposed can produce more accurate hand target contours, especially the five fingers.

Overall, after analyzing the extraction performance of the algorithms mentioned above , immune algorithm with minimum mean distance immune field gives the best segmentation result for target weakened traces infrared images.

Segmentation of target weakened traces infrared images is a new theoretical and application problem for image processing and criminal investigation. In this paper, we propose a new immune algorithm with minimum mean distance immune field for target weakened infrared image segmentation. This method use multi-step classification algorithm to achieve optimal classification based on the overall statistical properties of target areas and background areas. Besides, the immune variation and adaptive immune minimum mean distance recognition is presented, which will make sure that the extraction method would be appropriate for the target weakened traces infrared images.

The visual results using our proposed immune algorithm with minimum mean distance are compared with those using other image segmentation methods including classical edge detection operator, region growing algorithm, and watershed algorithm. The comparison results show that the proposed method can get good extraction results, especially the boundaries of five fingers, on target weakened traces infrared images.